Rezza maintains that he has submitted the requested documents and points out that both the ordinary courts and the public accounts court have already intervened on various matters. He is not asking to be exempt from scrutiny: he is asking for the respect due to anyone who has worked during a dramatic emergency, without for that reason being regarded as infallible. Then there is a question we risk forgetting: are we prepared for a new pandemic? Have we strengthened prevention, public health, epidemiology and research? Have we learnt anything from our mistakes? This is precisely what a parliamentary committee should also be addressing, because knowing who made mistakes yesterday serves, above all, to ensure we do not repeat the same mistakes tomorrow. His solidarity goes out to Rezza and to all those who have contributed, or could have contributed, to piecing together the facts. Science has no party affiliation, and expertise is not an aggravating factor. The doctors who tackled Covid are not public enemies. We owe the truth about the past to the public; we owe preparedness for the future to the next generations.

Matteo Bassetti, Full Professor of Infectious Diseases at the University of Genoa

Alice Buonguerrieri, an MP for Fratelli d’Italia, devoted a significant portion of the hearing to the possibility that even vaccinated people could transmit the coronavirus. Rezza explained repeatedly that vaccines reduced the likelihood of infection without eliminating it entirely, and that this reduction varied over time and with viral variants. The MP continued to ask what the scientific basis was for the restrictions on the unvaccinated, returning to the comparison with those who had recovered or were undergoing testing. The proportionality of the restrictions is a legitimate question, but one that requires an assessment of the expected health benefits and the alternatives available at the time they were adopted. It is not enough to demonstrate that a vaccinated person could transmit the virus to conclude that vaccination does not reduce transmission. Rezza even acknowledged that, according to current knowledge, the protection conferred by natural infection and that conferred by vaccination are roughly equivalent in duration. Not even this concession put an end to the dispute, as the MP continued to frame the debate in terms of justifying the restrictions, without quantitatively addressing the reduction in risk that Rezza had outlined.

Even more revealing was the discussion regarding the AstraZeneca open days. Rezza distinguished between the ministerial recommendations and the decisions of the regional authorities who actually organised those events, suggesting that the Commission should also hear from the regional officials. Buonguerrieri’s response was: “No, Doctor, on the contrary, we must hear from you.” The MP continued to attribute a political role to him which he disputed, repeatedly urging him not to get worked up and accusing him of failing to answer. In an investigation into accountability, establishing who made a decision is essential to determining to whom responsibility should be attributed. Here, when the person being questioned pointed to other institutional bodies that should be heard, his explanation was rejected and the personal criticism continued. A reconstruction of accountability that fails to examine the entire decision-making chain risks attributing to the central level even decisions made elsewhere.

Senator Antonella Zedda adopted a similar approach when questioning Rezza about conversations he had had in 2020 with Guido Silvestri regarding monoclonal antibodies. Faced with recollections that did not entirely match, dating back six years, she went so far as to ask him whether Silvestri was lying or whether he was suffering from ‘severe amnesia’. Yet Article 16, paragraph 4, of the Rules of Procedure allows the Committee to organise cross-examinations between individuals who have already been heard. There was therefore a means of clarifying the discrepancy through further investigation; instead, during the hearing, an alternative choice between lying and memory loss was repeatedly put forward. Zedda herself read out a communication dated 2 November 2020 in which Rezza reported that Minister Speranza had said to go ahead with the monoclonal antibodies initiative. The following day, the senator publicly claimed that Speranza was aware of Eli Lilly’s offer of free supplies and that the government had rejected it, without referring to that communication and without having fully explained the reasons why the clinical trial had not gone ahead.

The debate on home care followed a similar course. Buonguerrieri reiterated the ‘Paracetamol and watchful waiting’ approach, whilst Rezza referred to the text of the circular dated 30 November 2020, which also provided for the use of anti-inflammatory drugs. Subsequent studies were also cited to challenge recommendations made before that evidence was available. Responding to a question from PD MP Gian Antonio Girelli, Rezza observed that judging 2020 using information from 2021 and 2022 is “entirely misleading”. Determining whether an error occurred requires reconstructing the evidence available at the time of the decision, as this is the only way to establish whether it was avoidable. When Rezza attempted to protest, the chair intervened, effectively silencing him. Whilst the chair may certainly reprimand the person being questioned, it is the chair’s responsibility to ensure the fairness of the entire hearing.

It is precisely the conduct of the meeting that presents the most serious institutional problem. After Buonguerrieri had accused Rezza of treating dissenting scientists as “the village idiot” and had spoken of a “little charade”, the chair intervened to reprimand the witness, who was told that he could not continue to respond in that manner to a commissioner. Immediately afterwards, according to the official report, the hearing was suspended for four minutes. Later on, ‘quick questions, precise answers’ were requested. The committee members made personal attacks, whilst the reprimand was directed at those on the receiving end. This asymmetry is documented by the sequence of interventions. With the opposition virtually absent – only PD MP Girelli having spoken – the chair’s oversight of the fairness of the debate took on even greater importance.

The rules on this point are explicit. Article 82 of the Constitution grants committees of inquiry the same powers and subject them to the same limitations as the judiciary. Article 6(1)(b) of the Covid Committee’s rules of procedure entrusts the chair with the task of regulating discussions, whilst Article 16(3) expressly assigns to the chair the task of assessing the admissibility of questions. The Vice-Chair, Francesco Maria Salvatore Ciancitto, who chaired the entire sitting – as confirmed by the official summary report No. 153 of 10 November 2026 – therefore had a specific regulatory responsibility. The repetition of personal insinuations and the treatment meted out to Rezza’s responses raise a concrete question as to how that responsibility was exercised. Not every inappropriate question constitutes a breach; but a conduct that allows such allegations to be repeated calls for an explanation.

The conclusion of the hearing illustrates just how far the debate had gone. Zedda challenged the management of the risk of coronavirus transmission via aerosols, a serious scientific issue that does indeed merit a thorough investigation. During his speech, however, he raised the choice between “culpable negligence or a deliberate decision”, argued that there had been “stubborn persistence” in the wrong direction, and concluded by stating that this was not a question and did not merit a reply. Immediately afterwards, the chair closed the hearing. An allegation of liability was thus made publicly without giving the person concerned the opportunity to respond at that hearing.

The Constitutional Court, in its ruling 231 of 1975, clarified that the role of parliamentary committees of inquiry “is not to ‘judge’”, but “merely to gather information and data necessary for the exercise of the Chambers’ functions”. On Tuesday evening, however, the commissioners repeatedly passed judgement on personal responsibility while they were still collecting the statements needed to verify these claims. An inquiry must be able to amend its conclusions in the light of the responses received.

The public nature of the hearing confirms the problem. On 11 November, Buonguerrieri claimed that Rezza had admitted that all the government’s measures had “the sole aim of achieving mass vaccination”, announcing the collapse of the scientific justifications for lockdowns and the Green Pass. In the transcript, however, Rezza had stated that the government had promoted vaccinations “not only to save lives” but also to speed up the reopening of the economy. The difference is substantial: the MP’s statement omits precisely the public health objective explicitly stated by the epidemiologist. And acknowledging that a measure was adopted by the government through a political decision does not prove that that decision lacked a scientific basis. It is a flawed inference, constructed through an incomplete representation of the witness’s words.

The Commission’s mandate remains in place, enabling an assessment of what this lengthy session has actually achieved. Article 1 of Law No. 22 of 5 March 2024 stipulates that the inquiry must evaluate the measures adopted during the pandemic ‘also with a view to addressing a possible future pandemic of similar scale and severity’. Rezza had devoted a significant part of his report to the issues still outstanding: overcrowded A&E departments, community care homes not yet fully operational, delays in telemedicine and shortcomings in the surveillance of respiratory viruses, exacerbated by the lack of implementing decrees necessary for the new information system. The transcript does not reveal any substantive questions from the commissioners on these issues. After nearly four hours of actual proceedings, no one had delved into the vulnerabilities that Rezza had identified as priorities for tackling a future pandemic.

What happened on 10 November cannot therefore be dismissed as a mere clash between MPs and an expert unwilling to accept their objections. Article 7, paragraph 3, of the Rules of Procedure stipulates that the Bureau, supplemented by representatives of the parliamentary groups, shall examine procedural issues arising during the proceedings. This is the specific forum in which to review the recording of the sitting and verify how the powers conferred on the Chair by Articles 6 and 16 have been exercised. Should any breaches of the Rules of Procedure come to light, they must be acknowledged and addressed.

Members of Parliament are representatives of the Republic, vested with powers derived from the Constitution and the law. They are not lords who are permitted to exercise those powers at their own discretion. The safeguards afforded to elected representatives protect the freedom of the parliamentary function and coexist with the rules governing its exercise. Here, there are documented facts, identifiable lines of responsibility and precise regulatory provisions: the grounds therefore exist for demanding an institutional response.

The Commission should report on how the hearing was conducted and what useful insights actually emerged from it. Because the next pandemic will not be tackled more effectively thanks to the admissions the commissioners managed to elicit, but thanks to the genuine mistakes they were able to demonstrate.

Reading the article, one finally discovers the origin of the rumour that the average human attention span today is said to be around nine seconds – tragically shorter than that of a goldfish. The figure is said to have emerged from a marketing report by Microsoft Canada in 2015. Incidentally, it also appears that goldfish can retain memories for months.

However, to avoid referring to data that appears to be out of touch with reality, Adam cites the work of Gloria Mark, a psychologist at the University of California who, for twenty years, has been monitoring how office workers use computers – specifically, the time they spend in front of the screen carrying out a single task before moving on to the next. It emerged that during the early 2000s, the time spent at a computer on a single task was two and a half minutes; by around 2010, this had fallen to seventy-five seconds, and to forty-seven seconds in the early 2020s. Given that the reason for this susceptibility to distraction is thought to lie in the constant stream of rewarding distractions we are presented with – news, notifications, personalised shopping suggestions – and considering that, in laboratory tests, attention span does not appear to be generally impaired, the crux of the matter seems to revolve around our daily habits. If, throughout the day, one is constantly distracted by fast-paced, superficial and rewarding stimuli, one will struggle to read a book. However, if measures are taken to avoid these distractions and cultivate concentration, it is thought that sustained attention may return.

To avoid getting carried away, it is worth remembering that several studies agree that problematic internet use is associated with a reduction in grey matter in various areas of the brain. Indeed, Nilli Lavie, a cognitive neuroscientist at University College London, has shown that ‘grey matter volume can be used to accurately predict (attentional) performance’. Furthermore, it has been found that people with a greater volume of grey matter in the prefrontal regions are better able to resist distractions and maintain their concentration for longer.

Although the roles of genetics and experience in the relationship between grey matter and attention are not yet clear, when in doubt, it is best to try not to get too distracted.

These devices use electroencephalography, or EEG: electrodes on the scalp record variations in electrical potential produced by the activity of large populations of neurons. Commercial devices provide signals that are far less detailed than those from clinical equipment and do not freely read thoughts or memories. However, they can feed into algorithms capable of recognising sleep patterns or estimating certain cognitive states. The sensitivity of these recordings also depends on what future algorithms will be able to extract from them.

Moens examined the privacy policies of 57 companies marketing these neurotechnologies. Four had none at all, and eleven only provided statements relating to website visitors, without explaining how brain signals were processed. Among the companies with privacy policies, all but one permitted some form of sharing of aggregated data or data stripped of personal identifiers with third parties, without always clarifying whether EEG recordings were included. Only six expressly ruled out the sale of brain data. This does not prove that actual sales took place, but it does illustrate just how unclear the fate of such sensitive information was.

Interaxon, the Canadian manufacturer of the Muse headband, told Science that it had trained a model using approximately 50,000 hours of EEG data, most of which came from users of the product. According to the company, names and email addresses had been removed from the recordings. However, the privacy notice did not explicitly state that the signals would be used to train a general model of the brain. Following questions from the journalist, Interaxon amended it, acknowledging the need to inform users. A key aspect of the commercial agreement was only clarified following an independent investigation, after the training had already taken place.

Emotiv has also explained that it uses, for its Large Brain Model, some of the recordings from users of advanced devices, alongside other sources. Following enquiries from Science, it updated its terms and conditions to require explicit consent for training with pseudonymised neural data. This change confirms that using a device and contributing to the development of a commercial AI product are distinct purposes. Users must be able to authorise one without being obliged to accept the other.

To understand the economic value of the operation, one must consider the brain’s ‘foundational models’. These are systems trained on large quantities of data to learn statistical regularities and then adapted to specific problems, ranging from the classification of sleep stages to the recognition of certain neurological abnormalities. The competitive advantage also stems from the exclusive possession of signal collections that are difficult to reproduce. Manufacturers can build these collections using devices paid for by consumers, who are often unaware that their daily use contributes to creating a transferable and monetisable industrial resource.

Removing the name from the file does not automatically resolve the issue. Pseudonymisation means separating data from direct identifiers; anonymisation means making it not reasonably possible to trace the data back to the individual. These are different conditions. EEG tracings contain individual characteristics that are also studied as potential biometric identification tools. An algorithm can also derive information about cognitive states and other sensitive characteristics. Patrick Magee, Marcello Ienca and Nita Farahany explained in Neuron that mental privacy requires consideration not only of the original recordings but also of the inferences drawn from the data. Simply removing first and last names therefore does not provide sufficient safeguards.

There is also a technical asymmetry. Deleting a record from an archive is a traceable operation; removing the contribution it has made to the training of a neural network is far more difficult and not always verifiable. The right to withdraw consent is therefore at risk of being exercised only after the company has already incorporated the results of data usage into the product. Anyone who stops using the device may still have contributed to a model that remains usable.

The problem arises in an even more serious form in the workplace. A second report in Science details the use of helmets and other EEG devices to detect fatigue among workers and drivers, particularly in China. Preventing accidents is a legitimate aim, but the consent given by an employee subject to workplace surveillance lacks the freedom of that of a consumer who can choose not to use such a device. If the recording is linked to the worker’s identity and passed on to superiors or the authorities, the assessment of fatigue can turn into constant monitoring of the individual. Even errors in the algorithm can have professional consequences.

Europe has the General Data Protection Regulation, which requires specific purposes, transparency and special safeguards for health and biometric data where the conditions laid down by law are met. Not every EEG automatically falls within the same legal category, and the issue requires specific scrutiny of data collection and reuse practices. In the United States, Colorado and California have already extended legislative protections for neural data. General protections therefore require specific application.

The authorities should require a verifiable description of the data stored and its intended uses; separate, revocable consent that is not a prerequisite for using basic functions when commercial training is involved; and independent audits of training databases and transfers between companies. Stricter limits are needed in the workplace, as the right to opt out carries little weight when economic dependence comes into play. The origin of the data used by the models must also be verifiable.

The crucial issue is the distribution of power. Companies can transform data collected for a specific purpose into strategic assets, which can be exploited even when the company changes hands, whilst those who generated the data remain unaware of the full extent of the operation. The market rewards the accumulation of data and the success of models, but does not automatically account for the cost of future intrusions into individuals’ private lives. Entrusting companies with the task of defining the limits of what they can collect amounts to allowing those who profit from this expansion to determine the rules. Brain research can be carried out using legally collected data and independent oversight. What Science has documented, however, is a removal of decision-making power from those concerned, made possible by the weakness of public oversight. And it is precisely this that cannot be left to the market.

The story was pieced together by Jessica McKenzie in the Bulletin of the Atomic Scientists, thanks to a vast amount of private correspondence made public following legal proceedings. It is a particularly instructive case for understanding how artificial intelligence can alter the way scientific research is conducted, because it directly illustrates what happens when a tool capable of identifying the weaknesses in a piece of research is used by the very people who have a vested interest in defending its conclusions.

The report had been commissioned by the U.S. Secretary of Energy, Chris Wright, and entrusted to John Christy, Judith Curry, Steven Koonin, Ross McKitrick and Roy Spencer – scholars with genuine scientific expertise – who were chosen to produce a critical review of the current understanding of climate change. Published by the Department of Energy in July 2025, the document argued that many of the negative consequences of rising CO₂ levels had been exaggerated, whilst the potential benefits had been underestimated. The political stakes were high, as these conclusions could help justify the scrapping of federal regulations on greenhouse gas emissions.

A lawsuit brought by the Environmental Defence Fund and the Union of Concerned Scientists led to the discovery that the group had been formed in breach of federal regulations on transparency and the balance of advisory committees. In January 2026, a judge recognised the breach and ordered the documents to be handed over, allowing the publication of over one hundred thousand pages of material, including communications between the authors.

It emerges from these that Steven Koonin, a physicist and former chief scientist at BP, had begun experimenting with Microsoft Copilot as an assistant whilst preparing the report. After using it to summarise and rephrase certain passages, he asked it for a genuine critical assessment. Copilot noted that the publications cited, whilst often authentic and scientifically sound, did not adequately represent the full body of available knowledge. For example, the beneficial effects of CO₂ on plant growth were discussed whilst overlooking the constraints imposed by the availability of water and nutrients. The system also flagged a disproportionate reliance on publications by the report’s authors themselves and other researchers with similar viewpoints, accompanied by language that betrayed a polemical slant.

Copilot’s conclusion was explicit: the evidence was selected to support a favourable portrayal of the effects of CO₂. This is one of the most insidious distortions in scientific communication, as it allows a misleading conclusion to be drawn using genuine results. All that is required is to systematically exclude publications that contradict the desired thesis, whilst retaining those that are compatible. To uncover the problem, one must also be aware of the studies the author chose not to cite – a task in which artificial intelligence can be particularly useful.

The researchers’ initial reaction was enthusiastic. Judith Curry wrote that the result was surprising, even better than what might have been achieved by real scientists. Koonin said he was equally impressed. The two therefore recognised the value of the tool, but what emerges most clearly from the correspondence is the attempt to identify expressions that might have made the report appear prejudicially hostile to mainstream climatology. Koonin also asked Copilot to write an article opposing the document, in order to anticipate the critics’ arguments and incorporate pre-emptive wording into the final version.

This passage is crucial. The authors had received a complaint regarding their choice of scientific evidence, but the available documentation shows that they took particular care to make the text less open to criticism. This does not prove that they ignored all substantive comments, nor that they entrusted the drafting of the report to artificial intelligence – a claim they have explicitly denied. It does, however, show how easily a critical review can turn into an exercise in preparing for controversy, in which the primary aim is to anticipate and neutralise objections.

Copilot’s criticisms, of course, would have carried little weight had they remained mere judgements by a machine. Their validity must be established through independent verification of the claims. In September 2025, over 85 experts, coordinated by Andrew Dessler and Robert Kopp, published a comprehensive review of the government report, documenting numerous examples of selective citations and misrepresentations of the available evidence (Climate Experts’ Review of the DOE Climate Working Group Report, 2025). Subsequently, Benjamin Santer and four other climatologists demonstrated in Earth’s Future how the report had misrepresented key observations relating to stratospheric temperatures, using them to cast doubt on one of the pieces of evidence for human influence on the climate. They also examined a graph subsequently presented in public by Koonin, which contained erroneous observational data; the report’s authors later attributed this to a programming error (Misrepresentation of Human Climate Fingerprint in DOE Report, 2026).

The fact that artificial intelligence had recognised some of the inaccuracies in the document does not mean that its assessments were always correct. In one passage, for example, Copilot accepted the argument that it would be incorrect to speak of ocean acidification, given that seawater remains alkaline. This is a basic chemical error, because acidification means a decrease in pH, regardless of whether the final value remains above 7. The absorption of CO₂ causes precisely this decrease, as shown by observations collected by the National Oceanic and Atmospheric Administration. This incident demonstrates how essential it is to verify the models’ responses, even when their other observations prove to be well-founded.

Subsequent communications between the authors then reveal a growing impatience with the tendency of language models to favour the prevailing scientific literature. John Christy observed that a system trained on a vast amount of mediocre material would inevitably place too much value on that material. Roy Spencer suggested pressing the model with questions, pointing out that, with sufficient persistence, it is sometimes even possible to make a model change its mind. This is a valid observation regarding how these systems work, as they are sensitive to the wording of queries; however, it becomes problematic when the aim is to elicit a judgement more favourable to one’s own conclusions.

This brings us to the general significance of the matter. Artificial intelligence introduces an extraordinary possibility into scientific output: the ability to rapidly subject a piece of work to an extensive critical review, capable of comparing its claims with a vast body of literature and identifying possible omissions. The quality of this analysis naturally depends on effective access to sources and the system’s ability to interpret them correctly. However, once the document’s weaknesses have been identified, the same tool can be used to construct persuasive responses to objections without addressing the underlying causes that gave rise to them.

The consequences are likely to be very serious. Even today, we can produce texts that are scientifically flawed but extremely difficult to challenge on the merits, because any foreseeable criticism can be anticipated and met with a seemingly competent response. As the models improve, this possibility will grow even further. Scientific peer review thus risks becoming a contest between automated tools used to identify errors and equally powerful tools used to defend them, imposing an ever-increasing workload on human reviewers.

For this reason, artificial intelligence will only truly improve scientific integrity if its critical use is accompanied by procedures that ensure objections can be documented and responses verified. A well-founded objection should leave a verifiable trail, together with the sources supporting it and any amendments made to the work. It would thus be possible to distinguish a genuine scientific correction from a mere rewording that makes the problem less obvious. Furthermore, independent peer review must have access to this process, because leaving the management of criticism entirely to the authors means allowing them to select which criticisms they intend to take seriously.

Automation has been present in laboratories for decades, and as early as 2009, the robot scientist Adam was formulating and testing hypotheses on yeast genetics (https://www.science.org/doi/10.1126/science.1165620). Traditional instruments, however, simply carry out pre-set protocols: when a result suggests changing the experiment, a scientist must interpret it, design the new test and issue instructions to the various pieces of equipment.

The paper published on 4 November in Nature by Mingyu Wu and colleagues at Shanghai Jiao Tong University shows how this divide can be bridged (https://www.nature.com/articles/s41586-026-11228-6). AutoDNA makes the functions of over twenty types of instruments accessible to AI agents, which can be combined as required. Once given an objective in everyday language, the agents prepare protocols, instruct the machines and adapt the work to the results. The breakthrough lies in making the laboratory programmable by AI, which constructs and modifies the sequence of operations whilst working, linking one measurement to the choice of the next experiment.

The authors asked the system to optimise enzymatic DNA synthesis, which involves adding one base at a time to build a sequence. AutoDNA carried out successive cycles, modifying the conditions based on the observed yields, achieving up to 97.8 per cent per step, compared with the 92.7 per cent obtained when compared with a conventional statistical algorithm; it also coordinated four protocols in parallel, with execution 3.2 times faster than the sequence. It converted a message into 78 DNA fragments, synthesised and sequenced them, and recovered the text without errors after 11,382 operations in approximately eight and a half days. Finally, starting from a prepared tumour sample, it identified a previously known mutation in the KRAS gene and designed and synthesised the reagents to verify it experimentally: the result of the first analysis determined the operations for the second.

AutoDNA did not discover any new biological facts and worked on human targets using pre-configured tools; it requested supplies and maintenance, resolving eight of the fourteen recorded faults on its own. However, it demonstrated a fundamental component of the system that is still missing: an AI capable of formulating scientific hypotheses could use such a laboratory to select the evidence that distinguishes between competing explanations, read the measurements and autonomously review its own hypotheses. A researcher would no longer be required at every stage, although maintenance and quality control would remain essential.

If these systems were to become widespread – and there are economic reasons to believe they will – their cost-effectiveness would not depend solely on a reduction in manual labour: expensive tools could be used for longer and shared amongst many teams, whilst the automatic selection of experiments would reduce wasted attempts and make it feasible to explore combinations that are currently too numerous. The initial investment would remain high, but a widely used platform could reduce the cost of each new study, further accelerating adoption.

The most significant implication, however, concerns the relationship between what we know works and what we understand. Let us imagine that a platform identifies a combination of molecules capable of halting the growth of a tumour and that this effect is confirmed by independent evidence, right through to clinical trials. We might have an effective drug without being able to reconstruct the sequence of molecular interactions on which its benefit depends. Medicine is familiar with treatments whose effectiveness has preceded an understanding of the mechanism; the novelty would lie in making this gap systematic, producing usable results more quickly than we are able to explain them.

This gap would have practical consequences. If we do not sufficiently understand why a treatment works, it becomes more difficult to predict in which patients it will fail, what interactions it will have with other medicines and what long-term effects it may produce; specific experimental tests will then be required, rather than inferences based on a well-understood mechanism. An algorithm might also find a solution by exploiting particular patterns in the samples or measurements, which cease to hold true when conditions change. The reproducibility of the effect and the possibility of transferring it to new contexts would become central issues, all the more so if subsequent experiments were selected by another automated system.

Even the very meaning of scientific knowledge would risk changing. For a mathematician, a proof generated by AI may remain difficult to understand and yet still be formally verifiable; in biology, no single universal verifier can determine on its own whether a sample was representative or an instrument was correctly calibrated. The credibility of a discovery would require access to the original data and experimental decisions, as well as the replication of the experiments using independent methods. Even if the effect were confirmed, it would be necessary to derive an explanation from it that could be tested and used to predict new phenomena. A plausible account of the path followed by the AI would not guarantee that this explanation is correct.

The work of scientists would also change; they would have fewer opportunities to understand a phenomenon through the personal design and execution of each experiment and would have to devote more attention to selecting questions, verifying procedures and interpreting results. If the volume of data were to exceed human capacity to examine it, individual articles would no longer suffice to organise knowledge: verifiable archives and independent tools for scrutinising discoveries would be required. The training of scientists would also need to preserve the critical faculties necessary to evaluate results that none of them has produced directly.

The authors speak of ‘science as a service’, accessible even to those without a laboratory; however, this model could concentrate research capabilities in the hands of a few operators, who would decide which questions to explore and who can verify the answers. Independent verification would therefore require access to raw data and the ability to replicate the experiments outside the platform that conducted them, whilst rules on the ownership of results should prevent the mechanism of a therapy from becoming knowledge reserved solely for the owner of the system that discovered it.

AutoDNA is not yet a science capable of independently discovering new biological facts, but it lends credibility to the link between AI systems that formulate hypotheses and machines that test them against the physical world. If cost-effectiveness drives its widespread adoption, we may come to know an increasing number of phenomena through their effects, whilst our understanding of the underlying mechanisms will lag behind. The challenge will then be to build a science in which the speed of machines does not deprive humans of the opportunity to verify and understand what the machines have discovered, without ending up in a recursive process in which other machines are asked to interpret for humans what the machines have discovered and tested.

To understand why, one need only look at a mountain meadow. Many European meadows that we now regard as natural landscapes exist in their current form because, for centuries, people have brought animals there to graze or have cut the grass. Grazing limits the growth of shrubs; mowing prevents a few more competitive plants from dominating the others. The result is an open environment in which numerous species of flowers, insects and birds can coexist, often with very specific ecological requirements. When these activities cease, ecological succession begins: shrubs spread, trees take over the ground and the shade gradually eliminates many meadow species. The forest that forms can support other life, whilst species dependent on open spaces lose their habitat.

This phenomenon can be measured. In seven protected areas in the Czech Republic, a group of researchers compared historical and recent aerial photographs with data on the presence of butterflies and moths. With the expansion of forest cover and the disappearance of open habitats, the reserves have, on average, recorded a loss of 26 per cent of previously documented butterfly species and 19 per cent of moth species. This finding relates to areas that are already protected: designation as a reserve has not prevented an ecological transformation that is harmful to part of the fauna.

The issue directly affects Italy. In the Alpine valleys and inland areas of the Apennines, the abandonment of agro-pastoral practices is encouraging the encroachment of woodland and the closure of meadows. A 2026 study on the dry meadows of the Val di Susa reconstructed precisely this process, demonstrating the extent to which current biodiversity depends on the former extent of the pastures and the history of their use. The European Environment Agency also recognises the abandonment of species-rich meadows as one of the threats to specialised butterflies. Its meadow butterfly indicator fell by 47 per cent across the European Union between 1991 and 2024, due to the combined effect of agricultural intensification, habitat loss, abandonment and other factors. This figure stems from a number of causes, but it highlights the fragility of these ecosystems. A 2026 European Commission document estimates that one third of habitats of Community interest require active agricultural management, particularly extensive grazing.

The same logic applies in a very different landscape: that of Japan’s satoyama. These are mosaics of rice paddies, small traditionally managed woodlands, ponds and meadows, maintained for generations through the labour of rural communities. Their variety of habitats supports species that depend on periodic human activities, such as the irrigation of rice paddies or the cutting of vegetation. The ageing of the farming population and the abandonment of land are undermining these balances, to the extent that the OECD has highlighted their decline in its 2025 environmental review of Japan.

The effects of abandonment depend on the history of the area. Where agriculture has replaced native forests, spontaneous regeneration can restore valuable habitats. A study published in 2024 in Nature Sustainability, which examined 1,322 species of birds and mammals across eleven areas on four continents, estimated that the abandonment of cultivated land has increased the available habitat for 62.7 per cent of the birds and 77.7 per cent of the mammals considered. This represents significant evidence that allowing space for regeneration can be effective. Europe’s semi-natural grasslands and Japan’s traditional rice paddies offer further perspective: where entire biological communities have developed alongside sustainable human practices, discontinuing these practices means altering the conditions on which those communities depend.

The political issue, then, is far from marginal. If an area preserves its biological wealth through mowing, extensive grazing or the maintenance of small water systems, those who carry out these activities also fulfil a conservation role. A policy that merely funds the purchase of land to be taken out of human use may fail precisely where it aims to protect nature. In certain landscapes, it is necessary to provide financial support to those who continue to live in and manage them, recognising the ecological value of practices that the market, on its own, does not reward. This requires distinguishing between intensive agriculture – which is often responsible for severe losses of biodiversity – and extensive management, which preserves valuable habitats.

The fundamental error lies in imagining a single possible relationship between humans and nature, always expressed in terms of the degree of disturbance. Biology requires us to ask which species inhabit a given place, which processes enable their survival, and what happens when those processes change. In many regions of the world, human beings too have, over the course of centuries or millennia, become part of the ecological history of the landscape. Their disappearance may be a liberation for some species and a catastrophe for others. Conserving biodiversity requires an understanding of this history, even when it forces us to recognise that human presence may be a condition for its survival.

Yet, for over twenty years, this very institution has been at the centre of a staggering succession of scandals concerning the quality and authenticity of scientific publications, particularly those produced through its vast conference network. Thousands of articles have been retracted; conferences have featured non-existent or unreliable peer-review procedures; scientifically nonsensical papers have been accepted and published; and an organised market for the sale of fake scientific authorship has even emerged. The latest incident was reported by Nature on 30 October: the IEEE withdrew 1,417 articles in a single day, after an independent investigation had documented their links with commercial organisations specialising in the buying and selling of scientific publications (https://www.nature.com/articles/d41586-026-03455-8).

This affair should finally convince us to abandon the notion that the crisis of integrity in scientific publications is primarily a biomedical issue. The particular focus on fraud in medical and biological research – understandable given its potential consequences for public health – risks obscuring what is happening in the rest of the research community. In the case of the IEEE, the production of unreliable scientific publications has taken on industrial proportions within a system involving conference organisers, academic institutions and companies specialising in the fabrication of articles. The fact that such a prestigious organisation has taken decades to tackle the phenomenon seriously makes the affair all the more serious.

One of the first signs of this dates back to 2005, when three students from the Massachusetts Institute of Technology – Jeremy Stribling, Dan Aguayo and Max Krohn – developed SCIgen, a programme capable of automatically generating computer science articles by assembling technical terms, formulas, graphs and bibliographical references. The result looked like a normal scientific paper, whilst the content was completely nonsensical. One such article, entitled ‘Rooter: A Methodology for the Typical Unification of Access Points and Redundancy’, was accepted, without peer review, by the World Multiconference on Systemics, Cybernetics and Informatics, a conference supported by the IEEE. The experiment demonstrated just how easy it was to secure acceptance for a scientifically meaningless manuscript. The IEEE subsequently withdrew its support for the conference, and the case attracted considerable international attention (https://news.mit.edu/2015/how-three-mit-students-fooled-scientific-journals-0414).

In December 2008, an article produced using SCIgen and attributed to a non-existent researcher, Herbert Schlangemann, was accepted by the International Conference on Computer Science and Software Engineering in Wuhan, China, sponsored by the IEEE. The paper, entitled "Towards the Simulation of E-Commerce", even appeared in the IEEE Xplore digital library. The organisers invited its fictional author to chair a scientific session, based on an equally fabricated CV. The article was removed after the story became public. The experiment had demonstrated that it was possible to get non-existent results published by exploiting the lack of adequate scientific checks (https://diehimmelistschoen.blogspot.com/).

In 2014, the French researcher Cyril Labbé, who had developed tools to automatically recognise articles produced using SCIgen, identified over 120 nonsensical papers distributed by IEEE and Springer. A significant proportion of these originated from conferences organised in China between 2008 and 2013. The story made the pages of Nature once again, with an article documenting how automatically generated publications had entered the mainstream scientific literature. The problem had by then gone beyond the scope of a demonstration experiment: some people were using these publication systems to build up scientific CVs that could be used within their own institutions (https://www.nature.com/articles/nature.2014.14763).

In June 2015, Retraction Watch discovered that the IEEE had retracted thousands of conference papers, the majority of which had been published just a few years earlier. There were 1,281 retractions from the 2011 International Conference on E-Business and E-Government alone; a further 1,085 related to the International Conference on Bioinformatics and Biomedical Engineering held in the same year. The IEEE had also removed the proceedings of numerous other conferences from its digital library, deeming the procedures used to select the papers to be inadequate. The retraction notices, which were generally identical, cited generic breaches of the society’s editorial principles, without providing adequate individual explanations. Overall, a problem emerged involving thousands of articles and entire scientific events (https://retractionwatch.com/2015/06/25/one-publisher-appears-to-have-retracted-thousands-of-meeting-abstracts-yes-thousands/).

That episode could have been seen as an opportunity to radically overhaul the organisation of conferences. The succession of scandals demonstrates just how inadequate the measures taken have been. In 2018, the IEEE announced the retraction of 29 articles from the journal IEEE Transactions on Electromagnetic Compatibility, after ascertaining systematic breaches of its peer-review rules, and dismissed three editors involved. Twenty-six of the retracted articles were authored by the Italian engineer Antonio Orlandi, a former editor of the journal, without the IEEE publicly clarifying individual responsibilities regarding each paper. This was therefore an incident that also involved one of the society’s scientific journals and its internal editorial control mechanisms (https://retractionwatch.com/2018/11/22/engineering-prof-in-italy-earns-26-retractions-in-one-fell-swoop/).

In December 2022, the IEEE retracted over 400 papers from the 2016–2018 editions of the International Conference on Smart Cities and Systems Engineering, held in China, stating that it could no longer have confidence in the peer-review mechanisms adopted by the organisers. The measure affected entire conference proceedings, as the investigation had established that the procedures used to select the papers were inadequate (https://retractionwatch.com/2022/12/06/publisher-retracts-400-papers-at-once-for-violations-of-peer-review-process-policies/).

In June 2023, Retraction Watch documented a further set of anomalies in IEEE conferences, reported by Guillaume Cabanac and Kendra Albert. Among the published manuscripts were texts containing plagiarism, manipulated citations and scientific terms distorted by automatic paraphrasing programmes. In an article on the identification of lung cancer using machine learning, for example, a computational technique known as ‘random forest’ – which uses numerous decision trees to classify data – had been transformed into ‘arbitrary backwoods’, an expression devoid of technical meaning. A researcher interviewed anonymously also recounted how, at his institution, papers submitted to a conference sponsored by the IEEE could be assessed by colleagues from the same university, which was also the main source of the work to be published. The potential to inflate the institution’s publication figures and manipulate the organisers’ citation figures was evident. IEEE still maintained that its preventive controls were highly effective (https://retractionwatch.com/2023/06/15/plague-of-anomalies-in-conference-proceedings-hint-at-systemic-issues/).

In November 2024, Retraction Watch reported that the IEEE had retracted a further 57 articles from the journal IEEE Access, as they had been accepted without the minimum number of independent peer reviews required by the editorial guidelines. The articles had been published in 2020, and their removal therefore took place four years later. The same investigation documented how Cabanac had progressively identified an increasing number of suspicious publications in the journal using his own automated system for detecting textual anomalies (https://retractionwatch.com/2024/11/15/engineering-publisher-pulled-57-papers-in-a-day-for-peer-review-irregularities/).

A study published in 2025 in the Proceedings of the National Academy of Sciences, focusing on organisations that facilitate scientific fraud on an industrial scale, subsequently analysed approximately 2.3 million papers published at nearly 20,000 IEEE conferences since 2003, identifying 45 conferences held between 2009 and 2011 in which more than 10 per cent of the papers had been retracted. The researchers also identified conferences characterised by statistically anomalous concentrations of papers flagged on PubPeer – a platform where scientific publications are critically discussed – and of papers containing technical expressions altered by paraphrasing programmes. In at least seven conference series, the anomalies recurred for three or more consecutive years. These findings suggest that the problem may have persisted within the same organisational networks for years, without checks being able to put a stop to it (https://doi.org/10.1073/pnas.2420092122).

The scale of this publishing disaster is even more striking when one considers that, according to the study published in October 2026 by Anna Abalkina and colleagues, IEEE accounted for 34 per cent of retractions indexed in Scopus between 2001 and 2024. In the same study, a review of the Retraction Watch database as at 29 June 2026 attributed 15.5 per cent of all recorded retractions to IEEE, placing it in second place amongst publishers. These percentages are calculated from different archives and do not directly measure the proportion of fraudulent publications by IEEE, but they do document the exceptional position the organisation has assumed in the history of scientific retractions (https://doi.org/10.1080/08989621.2026.2733727). 

This series of incidents should suffice to demonstrate that the problem concerns the model for the production and distribution of scientific articles adopted by the IEEE for a significant part of its activities. The organisation sponsors more than two thousand conferences each year, producing around 350,000 conference papers. The most prestigious conferences, particularly in computer science, play a fundamental scientific role and may have extremely rigorous selection procedures. The IEEE brand, however, also encompasses a very different sphere, consisting of conferences organised locally by universities and associations, which may be granted technical sponsorship. In this system, the selection of manuscripts is entrusted primarily to the organising committees, whilst IEEE provides institutional recognition and, where papers are accepted, distribution via IEEE Xplore.

The distinction between scientific sponsorship and editorial responsibility has thus become particularly advantageous. Local organisers collect participation fees, authors secure a publication that is recognised in bibliographic databases, and IEEE expands its catalogue. There are various financial arrangements, including those in which IEEE units participate directly in any financial surpluses from conferences and those in which they receive fees for technical sponsorship. The official IEEE rules stipulate, for technically co-sponsored conferences without financial participation from the organisation, a fee of $1,450 plus $22 for each paper submitted for inclusion in IEEE Xplore. These sums are explicitly intended to cover the costs of the service and do not guarantee publication of the work. However, the same rules assign technical sponsors the responsibility of supervising the peer-review procedures: delegating the work to local committees, therefore, does not absolve them of responsibility for its quality (https://events.ieee.org/planning-basics/getting-started/obtaining-sponsorship/).

The financial figures are easily verifiable. In its audited financial statements for 2025, the IEEE reported $229.8 million in revenue from conferences, in addition to $321.5 million from periodicals, out of a total of $665.8 million in operating revenue. Direct costs for conferences in the same year amounted to $229.5 million, whilst in 2024 conference revenue had exceeded the costs attributed to that activity by more than $44 million. These figures demonstrate the sector’s significance within the organisation’s economy and the extent to which fluctuations in its financial performance are significant. The growth in the number of conferences and published papers nevertheless represents a commercial opportunity for the stakeholders involved in the system, whilst the investment required to properly evaluate each manuscript constitutes a cost. In the absence of adequate independent checks, the conflict between the volume of publications and scientific reliability is inherent in the very functioning of the system (https://2025.ieeeannualreport.org/wp-content/uploads/2026/06/Institute-of-Electrical-and-Electronics-Engineers-Inc.-2025-FS-FINAL.pdf).

Added to this commercial interest is that of institutions, which use the number of publications to evaluate their researchers. The phenomenon has taken on recognisable geographical characteristics, reflecting differences between local career systems and the incentives attached to scientific output. In India, where a particularly high proportion of the IEEE conferences covered by the investigations have been held in recent years, the regulations of many institutions place value on publication in proceedings indexed by Scopus and other databases, a key requirement for CVs, career progression and institutional evaluations. Some even provide cash rewards. The regulations that came into force in November 2024 at the Marri Laxman Reddy Institute of Technology and Management in Hyderabad, for example, provide for an incentive of 10,000 rupees for the first author and 8,000 rupees for the second author of a paper published in the proceedings of an international conference indexed by Scopus, subject to the conditions laid down by the institute. A contribution towards registration fees is also provided. These incentives are formally aimed at promoting high-quality research; however, they assign an easily measurable economic value to https://www.mlritm.ac.in/themes/custom/mlritm/includes/Policies/23.Final%20R%26D%20Policy%20-%2003-11-2024.pdf.

The result is a market in which the production of articles becomes profitable for parties with differing interests. A conference can boost its revenue by accepting more registrations; a department can improve its performance indicators by increasing the number of publications by its members; a researcher can buy a place amongst the authors, particularly when the cost of doing so is lower than the expected career benefit. Paper mills operate within this system by offering a commercial service that exploits the publisher’s prestige and the weakness of controls. The presence of many problematic papers at conferences organised in a particular region does not, of course, justify labelling research from that region as fraudulent. However, it does indicate a concentration of financial opportunities and institutional vulnerabilities that must be investigated and rectified.

The investigation that led to the October decision is particularly significant precisely because it maps out this market. Anna Abalkina, Marie Kunešová and their colleagues collected 4,407 advertisements offering positions to scientific authors, published on over 200 social media channels, including Telegram, WhatsApp and Facebook. By cross-referencing the titles on offer with the IEEE archives, they identified 2,062 articles, published between 2021 and 2026 in the proceedings of 366 conferences, which corresponded exactly or almost exactly to the papers offered for sale. The articles involved over 7,500 authors affiliated with more than 2,000 institutions across 56 countries. At some conferences, the identified papers accounted for up to 23.5 per cent of the publications. Of the 109 conferences with at least five articles linked to the advertisements, as many as 95 had been held in India (https://doi.org/10.1080/08989621.2026.2733727).

The prices help to illustrate how the scheme works. A first-author position could be offered for sums ranging from 43 to 400 dollars, whilst a last-author position could cost between 11 and 200 dollars. The correlation between the advertisements and the published articles is a very strong indication of the commercialisation of scientific authorship, although the mere presence of an offer does not prove that every transaction was completed or that every result contained in the article was fabricated. The falsification of authorship, however, constitutes a breach of scientific integrity. The authors of the study also documented anomalies in collaborations, the distribution of affiliations and citations, consistent with an organised system of commercial production and distribution of articles.

The IEEE responded by retracting 1,417 of those papers en masse on 23 October. At the time of the Nature article, 1,622 of the 2,062 papers identified by the investigation had already been retracted, whilst the total number of retractions carried out by the IEEE during 2026 was approaching 3,000. The speed with which an external report was translated into an editorial decision is encouraging. The IEEE had already carried out large-scale retraction operations in the past; the key difference this time concerns the identification of a commercial network offering scientific authorship and the scale of the response to evidence gathered by independent investigators (https://www.nature.com/articles/d41586-026-03455-8).

However, the quantitative defence put forward by the company remains baffling: it claims that the papers affected by the latest operation account for less than 0.06 per cent of all conference publications produced over the six-year period in question. This reasoning is statistically unsound: the papers were identified through a specific series of advertisements, which does not constitute a random sample of IEEE output and does not allow for an estimate of the overall frequency of fraud. Comparing the number of articles identified as problematic with millions of publications, in order to suggest that the phenomenon is negligible, confuses the ability to detect the problem with its prevalence. The thousands of retractions in previous years, the editorial checks that have repeatedly proved inadequate, and the existence of a public market for the sale of scientific authorship render this reassurance particularly implausible.

The October action is therefore a sign that deserves recognition, but the IEEE’s track record remains extremely poor. For over twenty years, there have been documented cases of gross fraud, conferences with unreliable peer-review procedures and entire batches of publications subsequently retracted. Meanwhile, these articles have had an impact on careers, academic rankings and subsequent publications that cited them. Retraction formally invalidates their scientific validity; it often comes after the benefits gained from those publications have already been reaped.

The IEEE’s responsibility therefore requires measures affecting the entire conference system: independent audits of scientific committees, traceability of peer reviews, preventive checks on authors and manuscripts, suspension of conferences that fail to demonstrate reliable procedures, and retrospective reviews of conferences where anomalous concentrations of publications are identified. The results of the investigations should be communicated to the authors’ and organisers’ institutions, so that the academic benefits obtained can also be re-examined. Organisations that award points and prizes simply for the presence of a paper in a database must bear some of the responsibility: an indexed publication merely documents that an article has been recorded, without certifying the validity of the research.

The decision to retract 1,417 articles in a single day shows that the IEEE finally possesses, at least in certain circumstances, the capacity to take action on a large scale and at short notice. The real test will be applying that capacity to routine conference proceedings, taking action before papers enter the literature and systematically evaluating those already published. After twenty years of scandals, the prestigious scientific society has yet to demonstrate that it can prioritise the quality of research—as its brand promises—over the commercial interests associated with conferences. The problem is not how quickly it can retract a thousand articles after someone else has uncovered the fraud, but how many millions of articles it has distributed without being able to guarantee that they had ever undergone scientific scrutiny worthy of the name.

Buonguerrieri, leader of the Fratelli d’Italia group on the committee, defended the provision allowing certain disbarred healthcare professionals to apply for reinstatement to the register, presenting it as the end of a ‘witch-hunt’ against doctors who ‘have saved lives’. He linked this to the committee’s work, which is tasked with bringing ‘the whole truth’ to light. However, the provision – which is proving controversial – does not provide for a blanket reinstatement: it sets out precise conditions, linked to the disputed facts and the circumstances of the emergency, and merely paves the way for an application to the Central Commission for Healthcare Professionals, to be assessed on a case-by-case basis. The bill has passed its first reading in the Chamber of Deputies, still needs to be approved by the Senate, and does not provide for any compensation. To interpret it as a rehabilitation of those who opposed official guidelines is to ascribe a meaning to it that the text does not contain. It does, however, serve to curry favour with the anti-vaxxer movement. It is the pandemic viewed through an anti-vaxxer lens, with Fratelli d’Italia and the Lega at the forefront and Forza Italia taking a more backseat role.

It is thus the anti-vaxxers and the promoters of dubious home remedies who are being celebrated as heroes, not those who, amidst the uncertainty of those months, worked on the available evidence, updating it as research progressed.

Borghi was more explicit. To those who asked what purpose the committee serves if no one ends up in prison, he replied that, ‘unfortunately’, it is not up to MPs to send them to prison, but that if a magistrate wishes to proceed, so much the better. The parliamentary inquiry becomes the premise – not to say the hope – of judicial punishment.

The commission has the right and the duty to scrutinise the decisions taken, identify errors and determine who is responsible. The problem is an inquiry that starts from a pre-determined verdict: not to understand how the institutions dealt with an unprecedented emergency, but to prove that the health policies were wrong and that those who criticised them were right. 

Proof of this is the persistence of the slogan ‘paracetamol and watchful waiting’, reiterated again yesterday: a formula that is as effective as it is misleading, a caricature of ministerial guidelines that are far more nuanced. But circulars are read less often than social media posts, and a simplification repeated often enough becomes the perceived truth, rendering any verification unnecessary.

The relationship between politics and science remains turbulent. Science does not produce immutable truths: hypotheses are revised, treatments are updated. But it does not follow from this that every past decision is indefensible simply because it is open to challenge. Decisions must be judged against the knowledge available at the time they were made. In November 2020, decisions were based on the best evidence available at the time; the arrival of monoclonal antibodies and antivirals has expanded treatment options, but this does not prove that previous recommendations were wrong. Fallibilism acknowledges error; it does not erase the distinction between knowledge supported by evidence from a specific period and unsubstantiated conjecture. 

The risk, therefore, is not that the commission will investigate, but that it will use the inquiry to turn a scientific controversy into a trial of its opponents: an ideological verdict in search of evidence – in other words, the opposite of the truth it claims to be seeking. The aim is to institutionalise anti-science by rewriting history according to a biased set of criteria.

I had denounced this move on 23 July, in an open letter to the Minister of Health, Orazio Schillaci, published in Il Foglio, detailing the initiative by Fratelli d’Italia MP Alice Buonguerrieri, who had tabled the amendment on 3 March, and the subsequent supportive intervention by Undersecretary Marcello Gemmato, despite the negative opinion which, according to media reports, had been expressed by the Minister. I asked Schillaci to clarify whether he agreed with the decision to offer political backing to anti-vaccine claims and how it was possible for an undersecretary to support a proposal on which the minister himself had expressed a contrary view. Three months on, those questions remain unanswered and the motion has reached the floor for a vote.

The Chamber of Deputies has approved Article 8 of the draft bill on healthcare professions by 126 votes in favour and 113 against. A section of Forza Italia abstained from the vote, whilst other MPs from the party contributed to the result; the eight MPs from Futuro Nazionale, the group linked to Roberto Vannacci, claimed that their support was decisive, and the figures confirm their assertion, because had those eight votes switched from ‘yes’ to ‘no’, the provision would have been rejected by 121 to 118. Fratelli d’Italia therefore championed the initiative, Forza Italia helped to secure its passage, and Vannacci’s supporters were able to present themselves to their own supporters as having been instrumental in its approval. It is difficult to imagine a more eloquent illustration of how an anti-vaccine stance can become a bargaining chip in the internal political competition within the right.

It is worth recalling what was actually approved, because confusion was, in fact, one of the tools used in this operation. During the pandemic, healthcare workers who failed to comply with the vaccination requirement were normally suspended from practising; striking off the register, on the other hand, was a much more serious disciplinary sanction, relating to further conduct deemed incompatible with professional duties. The provision passed by the Chamber of Deputies—which is still subject to completion of the legislative process—allows, under certain conditions, for an exceptional application to be made for reinstatement in the register for non-malicious acts committed during the state of emergency, from 31 January 2020 to 31 March 2022, with the assessment of such applications entrusted to the Central Commission for Healthcare Professionals. No reinstatement is automatic. However, there remains a question that the proponents of the legislation should answer before seeking to amend the disciplinary system: what mistakes did the professional bodies make, how many professionals were affected, and why would the normal appeal and reinstatement procedures be insufficient? To justify exceptional legislation, documented cases and verifiable reasons are required, which so far have not been at the heart of the public defence of the measure.

In their place came statements from Buonguerrieri, who spoke of persecution and a “witch hunt”, and from Lucio Malan, chair of the Fratelli d’Italia senators, who accused opponents of the legislation of longing for a time of hatred and division, whilst questioning the scientific basis of the measures adopted during the pandemic. These are extremely serious accusations against the health authorities, made by representatives of a governing party who would have all the means to verify their validity. If Buonguerrieri is aware of cases of disciplinary persecution, he should document them; if Malan believes that the health measures were scientifically unfounded, he should specify which ones, on the basis of what evidence and in relation to what knowledge was available at the time. Their institutional role makes the casual use of accusations that fuel mistrust of medicine and of those tasked with protecting patients particularly serious.

This is the point on which we must hold political leaders to account. Even small anti-vaccine groups can guarantee a concentrated, vocal and easily mobilised electoral base, which is particularly attractive to parties and candidates seeking to distinguish themselves from their allies. The statements made by the campaign’s organisers and the Vannacci supporters’ claim to have secured the decisive votes demonstrate just how useful this battle has become for gaining political visibility. There is no need to know the private intentions of MPs to judge the choice they have made: they supported an exceptional measure, accompanied it with a narrative of persecution against dissenting healthcare workers, and offered anti-vaccine groups a political victory that they can legitimately claim as their own. The potential electoral gain is immediate, whilst the cost of weakened confidence in vaccinations falls on the entire population.

That cost is well known. Vaccine misinformation influences decisions on which the protection of children, the prevention of infectious diseases and the health of the most vulnerable people depend; when it is spread by a doctor, their professional authority can make it even more persuasive. Professional bodies also exist to ensure that professional titles are not used in breach of ethical duties, and anyone wishing to challenge their decisions has a duty to demonstrate which injustices they intend to rectify. Portraying those decisions as persecution, without providing evidence, amounts to providing fresh ammunition to those who claim that health authorities have deliberately misled or punished the public. Parliament should be aware of the consequences of such a message, particularly in the wake of a pandemic in which misinformation has hindered prevention and contributed to harmful health decisions.

The coincidence with Mattarella’s speech at AIRC makes the irresponsibility of this move even more evident. The immunological knowledge that has enabled the development of vaccines also underpins new therapeutic strategies against cancer, including experimental applications of messenger RNA platforms. While research is opening up treatment possibilities that were hard to imagine just a few years ago, members of the ruling coalition are fuelling a narrative that discredits the scientific institutions on which the credibility of these innovations also depends. The trust required for citizens to accept effective prevention or treatment takes years of work and consistent institutional behaviour to build; yet it takes only a few irresponsible statements – especially when they come from those in government – to reinforce doubts that disinformation has already spread.

The greatest responsibility lies with Giorgia Meloni, because the initiative originated within her party, is supported by an undersecretary in her government and defended by the president of the Fratelli d’Italia senators. The Prime Minister should explain whether she agrees with the accusations of persecution levelled at the health authorities and whether she considers the vaccination measures adopted during the pandemic to be scientifically unjustified. Above all, she should clarify whether she considers it acceptable for her own party to seek political support from anti-vaccine groups by offering them a legislative and symbolic victory, whilst the National Health Service faces far more urgent problems, ranging from staff shortages to inequalities in access to care. Forza Italia is accountable for the votes that enabled the bill’s passage, despite the reservations expressed by some of its MPs; Futuro Nazionale is accountable for the support it has proudly claimed. Those who backed the legislation must specify what benefit to patients justifies the exception introduced and why it should take precedence over health issues of far greater significance.

Schillaci remains; I shall repeat to him the questions I put to him in July. Is it true that the Ministry had expressed a negative opinion on the amendment? Who authorised Gemmato to support it in the Committee, and what official position has the Ministry adopted? A Health Minister who is also a doctor and a man of science understands the value of trust in vaccinations and the role of disciplinary accountability; for this very reason, he must explain how he intends to reconcile his role with a political initiative which, through the statements of its promoters, lends legitimacy to the narrative of anti-vaxxer persecution. Public silence on such a serious disagreement, should the journalistic account be confirmed, constitutes a political problem in itself.

A democracy also recognises the right of anti-vaxxers to organise, demonstrate and seek representation, but those in government have a duty to assess their demands on the basis of the public interest, and not on their ability to win votes. It is precisely this duty that the key figures in this affair seem to have subordinated to political expediency, choosing to give legislative weight to a demand accompanied by inadequately substantiated accusations against health institutions. The most insidious consequence concerns the precedent that is set: if a sufficiently mobilised minority manages to secure exceptional treatment by portraying the application of professional rules as persecution, other groups may attempt to follow the same path, thereby turning scientific and ethical responsibility into a matter of electoral bargaining.

Mattarella pointed out that a priori opposition to vaccines can cost lives, and none of those involved in this operation can claim to be unaware of the danger. Meloni, Buonguerrieri, Malan, Gemmato – the Forza Italia MPs who voted in favour and those from Futuro Nazionale who asserted their role – have had months to reflect on the consequences of their choices, whilst Schillaci has had ample time to clarify the Ministry’s position. They must now publicly account for the health and scientific grounds that would justify this measure, for the allegations levelled at the disciplinary bodies, and for the decision to cultivate political support based on mistrust of medical institutions.

The advantage of this tactic is also evident in the distribution of responsibility: any votes gained are immediate and easily attributable to the politicians who supported the legislation, whilst the consequences of health misinformation may only emerge years later and affect people who will never know which statements contributed to their beliefs. It is one of the most irresponsible forms of electoral opportunism, because it allows one to reap a political benefit whilst leaving the community to bear the risks of one’s choices. Anyone who believes that a few thousand anti-vaccine votes are worth this price should at least have the courage to declare this to the Italian people: because the votes they hope to gain belong to them, whilst the health they are putting at risk belongs to everyone.

Shuford was one of the few figures in whom public health experts had placed some hope. An epidemiologist and former Texas health commissioner, she had led the state’s response to the 2025 measles outbreak, earning a reputation as a competent professional, data-driven and aware of the importance of childhood vaccinations. Her arrival at the CDC last June had been welcomed as a sign that, after months of upheaval, some experienced professionals would remain to tackle the diseases affecting Americans. In her official statement, Shuford thanked Trump and Kennedy for the extraordinary experience and explained that she wished to devote more time to her family. Personal reasons must be respected. However, the Washington Post also reports that there is disagreement over the administration’s vaccination policy, according to a well-informed source. What the decisive reason for her departure was remains unclear at this stage. The result remains the same: another qualified professional is leaving a top post at an agency that is in desperate need of continuity.

To fully appreciate the significance of this news, it is worth recalling what has happened at the CDC since Trump took office. The first nominee for the post, Dave Weldon, had his nomination withdrawn before the Senate confirmation vote. He was succeeded by Susan Monarez, a scientist who had previously served as acting director of the agency and who, once confirmed, lasted less than a month. In August 2025, she was sacked following a clash with Kennedy, to whom she had refused to grant carte blanche over vaccination decisions and the management of scientific staff. Her dismissal led to the resignation of other senior executives, including Debra Houry, the CDC’s chief medical officer; Demetre Daskalakis, director of the Centre for Vaccines and Respiratory Diseases; and Daniel Jernigan, head of emerging infectious diseases. These professionals, who had spent years building up expertise, surveillance programmes and epidemic response capabilities, had left, denouncing the politicisation of public health.

Following Monarez’s departure, an interim editor-in-chief, Jim O’Neill, was appointed, and eventually Erica Schwartz took over, alongside a new management team that included Shuford. The change of leadership was supposed to resolve the crisis. Today we find ourselves once again in the position of seeking a replacement, with Satish Pillai tasked with temporarily taking over the duties of the outgoing director. With every change, the names on the office doors change, whilst the scientific expertise accumulated by the institution continues to be lost. Above all, it is becoming increasingly difficult to imagine what a competent director could achieve when the decisions they are supposed to make on the basis of scientific evidence depend on the political leadership’s willingness to accept them.

The measles outbreak highlights the consequences of this approach to governance. After decades of successful vaccination campaigns, the United States is witnessing the return of a disease that it had declared eliminated from the country in 2000. As of 22 October, confirmed cases in 2026 had reached 4,410, almost double the number recorded in the whole of 2025. In Pennsylvania, where a particularly severe outbreak has developed, state health authorities have reported five deaths linked to measles, whilst the CDC recognised just three across the whole country. The discrepancy also concerns the method of classifying deaths: the federal agency changed its counting system, relying on death certificates rather than data transmitted directly by the states, with consequences for the timeliness and completeness of the available information. According to a Reuters investigation, Kennedy is said to have personally asked the new director, Schwartz, to remove certain deaths reported by Pennsylvania from the federal tally. The Department of Health has disputed that account. One extraordinary fact remains: whilst measles is spreading, federal and state authorities are publicly arguing over the number of people who have died from it.

Yet, for measles, we have one of the most effective vaccines ever developed. With two doses, protection against the disease reaches around 97 per cent. We know the vaccination coverage thresholds required to prevent the virus from circulating, and we know which population groups are most at risk of complications. The problem in the US now concerns the ability of institutions to use established scientific knowledge to protect the population. Kennedy, who built much of his reputation by spreading unfounded doubts about vaccine safety, now heads the federal department that is supposed to promote their use. And Trump continues to support initiatives that follow the same line, including the proposal to split certain combined paediatric vaccines into multiple doses, without any proven general clinical need to do so.

The institutional outcome is equally significant. In June 2025, Kennedy simultaneously dismissed all seventeen members of the Advisory Committee on Immunisation Practices – the expert committee that formulates vaccination recommendations for the CDC – replacing them with people of his own choosing. In January 2026, the administration amended the federal paediatric vaccination schedule, reducing the number of universally recommended vaccinations. A federal court subsequently suspended the implementation of those changes and the effects of the committee appointments, giving rise to further legal proceedings. Thus, the uncertainties introduced by political decisions were compounded by those arising from legal challenges. Scientific societies and several states have begun to draw up guidelines independent of the federal ones. One of the main advantages that a national health system can offer is gradually being lost: the availability of consistent recommendations, developed through shared scientific procedures and stable enough to guide the work of doctors.

The disaster, of course, also affects biomedical research. The National Institutes of Health, which for decades have been one of the main drivers of global medical research, have been hit by a campaign to cut funding, halt projects and reduce staff numbers. By January 2026, according to a report published by Nature, the administration had already cut or frozen over 7,800 research grants across various federal bodies, whilst around 25,000 staff had left the scientific agencies. Further investigations carried out in the following months documented the impact on universities, laboratories and researchers involved in multi-year programmes. Some of the measures were overturned by the courts, but the restoration of funding does not automatically make up for lost time, nor does it guarantee that a disbanded research group can be reconstituted.

Here, an aspect of Trump’s policies emerges that deserves particular attention. Scientific research is a long-term endeavour, requiring coherent programmes, specialist expertise and a reasonable degree of predictability in working conditions. A laboratory may take years to recruit the necessary staff and develop experimental technology. An epidemiological surveillance programme derives its value from the continuity of its observations, as this is the only way to recognise a change in the incidence of a disease in a timely manner. The federal government is introducing a degree of instability into these activities that is undermining their functioning. Every researcher who loses funding, every epidemiologist who leaves an agency and every programme that is discontinued represents a loss of scientific capacity that will take time and resources to recover.

All this is taking place under the banner of the ‘Make America Healthy Again’ movement, through which Kennedy promises to restore health to Americans and free institutions from the influence of big pharmaceutical companies. The promise might seem appealing, were it not for the criteria used to select which areas of expertise to promote. Mistrust of scientific authorities has itself become a political credential, whilst a willingness to apply well-documented knowledge can become a reason for incompatibility with the administration’s leadership. Conflicts of interest within the pharmaceutical industry are a real problem, requiring rigorous oversight and transparency. Entrusting health policy decisions to a secretary who has repeatedly promoted claims contradicted by scientific evidence offers no guarantee of greater independence.

The point is that Trump and Kennedy seem to regard scientific institutions in the same light as any other administrative body over which they can exercise their power of appointment and political direction. This is a view that ignores the specific role of these institutions. A minister may legitimately decide on the general priorities for healthcare spending, whilst taking responsibility for these decisions before the electorate. The efficacy of a vaccine, the existence of an epidemic risk and the interpretation of clinical data, on the other hand, require verification procedures that must be capable of producing results which may be unwelcome to the minister himself. A health agency loses its usefulness when this possibility is subordinated to the political convenience of those who run it.

Jennifer Shuford is set to leave on 12 November. Perhaps it is indeed for family reasons, as she has stated, and it would be wrong to attribute intentions to her that she has not publicly expressed. But this latest departure makes an even clearer picture of a phenomenon that now goes beyond the motivations of the individuals involved. Qualified professionals who join the civil service to restore credibility to its decisions end up working within a system in which that credibility is eroded on a daily basis by the actions of the very government that appointed them. Some are sacked, others resign, whilst others remain and are forced to deal with the consequences of policies contested by a significant section of the scientific community.

It has taken the United States decades to build a biomedical research and public health system that has made a decisive contribution to medical progress worldwide. Trump and Kennedy are demonstrating just how quickly it is possible to undermine its functioning. The most disturbing thing is that the damage will continue to have consequences even after they have left the scene: researchers who have changed careers, laboratories that have closed down and eroded confidence in vaccinations will not return by decree. As for the CDC, a new deputy director will be sought, who will probably be presented as the right person to restore order and confidence. Then we shall see how long that lasts. Measles, meanwhile, does not have to go through any appointment or confirmation process, and quietly carries on doing its job.

Despite this – that is to say, despite the temptation to gloss over matters, to avoid facing up to them and to refrain from publicly exposing what occasionally comes to light, particularly when it concerns systemic failings which are therefore more damaging and harder to rectify – I believe that the ethics of knowledge outlined with crystal-clear clarity by the Nobel laureate Jacques Monod – that ‘duty to truth’ which entails a responsibility for every researcher, every scientist and, I would say, every intellectual in general to candidly expose their own errors and those of others, whoever may have discovered them.

Please note: ‘self-correction’ – that term invoked to demonstrate the superiority of the scientific method (or methods) over others – does not merely imply the correction of one’s own errors, but must be understood as a form of action undertaken by the entire scientific community: it is the entire scientific community – or rather, the majority within it – whose opinion must shift in the face of an error; self-correction is therefore a collective endeavour, in which everyone – through the famous ‘peer review’ process – has a duty to correct any errors that come to their attention, whether by chance or through research.

Not only that: this process, within the limits of what is feasible, must be transparent to the public, because only in this way can trust in science – and the justification for the superiority of its method – be upheld; highlighting errors, and showing how and when – that is, under what circumstances – they are corrected, is the best way to illustrate how and why progress is made in humanity’s understanding of the natural world.

That is why I was once again (as has happened many times in the past) unpleasantly surprised by certain comments prompted by my latest article on a serious issue of falsified data uncovered by Sholto David in the scientific literature.

“It’s nothing new”, “why are we surprised?”, “it’s been like this for at least 20 years” – this is the gist of such comments. When confronted with the obvious rejoinder that such comments are generic—and consequently with the request to explain what they are based on—it transpires that certain colleagues have not only been grappling for decades with the difficulties of replicating the data published by the basic research necessary, for example, to identify new targets for innovative therapies, but that there are even those who have devoted considerable effort to attempting laboratory replicas, identifying the problems, in short, in demonstrating that what has been published does not hold up. But then, one wonders, with those data in hand, and with that awareness of so many problems in the scientific literature, why did you not follow the example of David – to stick with the most recent case – or that of myself, who for some fifteen years has been correcting instances of errors, inaccuracies, fraud and whatever else I happen to encounter, in every forum and by every means at my disposal?

Because, I am told, amongst other things, “we’re not the police”.

It is true: we are not. But we are all – in industry and the general public, researchers and scientists alike – bound by that ethics and that responsibility I referred to, which Monod so aptly articulated, and these do not allow us to shirk an indispensable part of the scientist’s work: the duty to scrutinise, to critique, to uncover and to report potential errors, so that the entire community can examine a result that has flaws – or is presumed to have them – and determine how sound the published work is.

It cannot be the case that, simply because we work in industrial development, we are not interested in the state of scientific knowledge; nor can we afford to look the other way – a habit all too often adopted for the sake of convenience, to avoid causing a disturbance, to spare ourselves the effort, and to avoid debate – as if scientific literature or the publications of others were of no concern to us, except when they provide valid support for what we are doing.

Turning a blind eye, ‘because we aren’t paid for this’, is the scientific manifestation of that lack of civic spirit which is leading, more generally, to the dissolution of our societies; and precisely for this reason, I shall conclude by making my own a phrase by Gramsci from "La città futura", which, strangely enough, also applies within the academic and scientific community: I hate the indifferent. In the hope that the benevolent reader will excuse this impassioned plea of mine – perhaps a little too impassioned – I would remind everyone that in science, recognising an error – and perhaps even demonstrating it in the laboratory – without communicating it to others is no less culpable than producing a forgery and failing to remove it from the path of knowledge.

David has compiled 230 reports of protein expression that is incompatible with the known genetic characteristics of the cells studied, spread across 221 publications. The principle behind his investigation is very simple. When a gene has been completely removed from a cell’s genome, that cell cannot produce the corresponding protein. If a scientific article features a photograph in which the protein appears nonetheless, something in the experiment must necessarily be wrong. Yet those photographs were published, often accompanied by elaborate interpretations of the biological mechanisms they were supposed to demonstrate.

Let us consider a case dating back to 2008. MCF-7 cells, derived from a breast tumour and widely used in cancer research, have a 47-base deletion in the caspase-3 gene, a protein involved in programmed cell death. This alteration prevents the production of the normal protein, a fact documented as far back as 1998 in a paper published in the Journal of Biological Chemistry. However, an article published in the International Journal of Oncology in 2008 demonstrated, using a technique called Western blot, the presence of caspase-3 in those cells. Western blotting enables the detection of a protein using an antibody and produces a dark band at its location. In the case of MCF-7 cells, that band should not have been present.

One might think this was an isolated error. In 2015, a paper published in Anti-Cancer Drugs, focusing on a natural substance called zerumbone, described the effects of the treatment on the STAT3 protein in PC-3 prostate cancer cells. The problem is that these cells have a deletion encompassing the entire STAT3 gene, a fact already documented in 2003 in Oncogene. Nevertheless, the authors presented a Western blot indicating its presence, basing part of their interpretation of the observed pharmacological effect on that signal.

Clearly, we do not need generative artificial intelligence to produce such results. The examples from 2008 and 2015 predate by a long way the widespread use of current systems capable of generating plausible experimental images. The phenomenon has deeper roots and includes possible exchanges of cell lines, antibodies that recognise proteins other than those stated, experimental errors and the manipulation of results. David’s compilation identifies contradictions that require explanation; it would be incorrect to automatically attribute each one to deliberate fraud.

The result remains devastating. Peer review failed to identify contradictions that could have been verified by consulting already available genetic information. In another example discussed by David, an antibody used to recognise the BAX protein – cited in over 1,400 publications – was only identified as problematic in 2024, through a study published in Cell Death & Disease. This means that for years, the same reagent may have underpinned a vast number of scientific findings whose interpretation is now in doubt.

The problem becomes even more serious when such findings find their way into literature reviews. David demonstrates how studies on STAT3 signalling in PC-3 cells – which are incompatible with their genetic makeup – have been cited in review articles. At that point, the erroneous information begins to appear as though it has been confirmed by the convergence of various publications. Artificial intelligence systems that analyse the literature can reproduce and further amplify it. The proliferation of citations thus lends an appearance of validity to what was incorrect from the outset.

And yet David has identified only those errors that can be recognised thanks to a previously known biological incompatibility. How many false results remain undetected because they describe events that are biologically possible but have never actually occurred? How many experiments that were never carried out are indistinguishable from genuine experiments once they have been transformed into figures in an article? David’s sample, compiled through targeted research, does not allow us to estimate the percentage of erroneous publications. It does, however, demonstrate that even elementary contradictions can survive the scientific scrutiny system for years.

The success of fraud also depends on an asymmetry that will continue to worsen. To produce a false result, one need only alter a photograph, invent a set of numbers or present a non-existent experiment as one’s own. Recognising this requires specific expertise, access to the original data and, often, an investigation that takes weeks of work. When a technique capable of detecting manipulation is developed, those seeking to deceive can learn to circumvent it. It is an evolutionary race in which methods of falsification are selected precisely because of their ability to bypass the available checks. Artificial intelligence accelerates this process, reducing the cost of producing seemingly credible results.

Continuing to invest primarily in detecting fraud after publication means accepting that we are chasing an adversary who can constantly change their strategy. We need to intervene at an earlier stage, when the scientific data is being produced.

The necessary technology exists. A modern laboratory instrument generates digital information that can be automatically recorded, linked to the identity of the instrument and the experiment, time-stamped and protected using cryptographic procedures. A digital signature can certify the document’s origin, whilst a specific mathematical fingerprint, known as a hash, makes it possible to verify whether even a single element of the file has been subsequently altered. By recording these fingerprints in a distributed ledger – a blockchain – it becomes possible to construct a verifiable history of the data’s creation and any subsequent changes.

Technologies of this kind have already been studied and tested in clinical research. In 2022, Scrybe was unveiled – a system that uses a permissioned blockchain to record the provenance of clinical data, with a prototype integrated into the REDCap platform, which is widely used to collect trial data. The system is described in an article published in Digital Threats: Research and Practice. A paper published in December 2025 in Frontiers in Medicine described the implementation of an architecture based on Hyperledger Fabric, verified using data from a pilot clinical trial, to ensure the traceability of operations and information. These applications are still limited, but they demonstrate that the key technological tools are available.

Of course, certifying the origin of a piece of data does not automatically prove that the experiment was conducted correctly. Even a genuine instrument can measure the wrong sample, and a defective antibody can generate a perfectly genuine signal. The cryptographic guarantee concerns the provenance and integrity of the record, whilst the biological quality of the experiment requires appropriate controls. In the case of the Western blots examined by David, the genetic identification of the cell lines and the validation of the antibodies therefore remain essential. Data certification would become truly effective by linking this information to the instrumental results from the moment they are acquired.

We can therefore envisage a laboratory in which the digital original of every experiment is automatically recorded in an independent archive, together with the information needed to interpret it. A researcher could freely process it, improve the visualisation of an image, carry out statistical analyses, correct errors and repeat calculations. Every step would remain verifiable, as the initial data would be preserved and the transformations documented. The greatest advantage would be that it would make it much more difficult to retroactively replace unwelcome results with more convenient ones, without leaving a trace.

To implement a similar system, tools are required that can stamp data directly upon acquisition, reliably identify operators and samples, and provide independent archives and verification procedures. Blockchain would be one of the technologies that could be used to guarantee the immutability of records, without having to store images or entire datasets on it. Cryptography already makes it possible to separate data storage from proof of its integrity. Regulatory authorities, moreover, have long required original records and systems for tracking changes – so-called audit trails – in numerous pharmaceutical and clinical contexts. Extending these principles to routine scientific research would be a challenging change, but one that is perfectly compatible with current technological capabilities.

Such a transformation, however, requires us to address an even more fundamental question: what role should scientific publication play?

Today, an article published in a journal continues to represent, in practice, the primary unit of certified knowledge. It is the item that is cited, used in academic assessments, included in systematic reviews and analysed by artificial intelligence. Yet the article is essentially the authors’ account of their own experiments. The data presented are those they have selected and organised to support their conclusions. The publishing system verifies this account primarily through the judgement of other scientists, who are rarely provided with the tools necessary to reconstruct the entire experimental process.

We should begin to regard publication as the moment when interpretations of research are discussed, thereby granting experimental data an autonomous and verifiable existence. The fundamental heritage of science would thus consist of large, interoperable archives of original data, accompanied by information on their production, quality controls and available certifications. Verification of provenance and that of scientific validity would be explicitly distinguished, and a result could be re-evaluated without relying exclusively on the article in which it was presented.

A significant proportion of the results could also be subjected to automated checks even before they enter the scientific literature. It would be possible to compare the declared genetic identity of a cell line with the protein that an experiment claims to measure, to check the reagents used, and to verify the consistency between original data and published analyses. David’s work demonstrates precisely that information already available in various archives can be combined to identify contradictions that have escaped human detection.

The benefit of having certified scientific databases would become even greater in the age of artificial intelligence. These systems acquire knowledge from a body of literature that contains a mix of reliable findings, errors and fabrications. They are capable of rapidly processing volumes of literature that no scientist could possibly read, but the quality of their responses also depends on the quality of the information used. The case of the non-existent proteins demonstrates just how easily an error can be reproduced from one publication to another, to the point where it is assimilated by the automated systems we use to navigate our research.

The prospect of further growth in scientific output, fuelled by the automated generation of texts and images, makes it absolutely essential to distinguish between what has actually been measured and what someone has merely claimed to have measured.

For this reason, universities, research organisations, funding bodies and publishers should agree on common standards for certifying data at source and support public infrastructures capable of preserving it in the long term. Research funding should include the production of verifiable records, their preservation and the availability of the metadata essential for verifying them, with the necessary safeguards for confidential data. Journals could require a documented provenance of experimental evidence, clearly indicating which controls were carried out.

Such a transformation would require investment and a fundamental shift in academic practices. It would also make it possible to allocate an increasing proportion of scientific resources to the production of reliable knowledge, rather than expending them in an attempt to identify false information retrospectively.

Sholto David’s work should raise concerns that go far beyond the 221 articles identified. For at least twenty years, findings incompatible with basic biological knowledge have been published, whilst we continue to regard publication as the primary guarantee of a discovery’s reliability. Meanwhile, we have developed technologies capable of verifiably recording the origin of information, and today we have IT systems capable of automatically comparing vast amounts of data.

Continuing to build scientific knowledge without demanding guarantees regarding the provenance of the evidence underpinning it is an increasingly untenable choice. If we allow the production of fakes to grow faster than our ability to recognise them, we risk gradually losing the very ability to distinguish accumulated knowledge from its imitations. The time has come to change the rules, starting with what a tool has actually recorded, before anyone turns it into a publication.

As long as everyone is convinced that the future market will be large enough to recoup current investments, the system works perfectly well. However, there is one question that someone will have to ask sooner or later: how much money is coming in from end consumers – those who actually buy the products – and how much continues to circulate amongst companies that finance one another?

This is a particularly interesting question when applied to artificial intelligence, as it describes quite accurately some of the financial relationships that currently link some of the sector’s leading companies. The mechanism is known as circular financing, and an article published on 18 October in The Conversation by Ruipeng Liu and Sagarika Mishra, entitled ‘Who’s paying whom? The circular money machine behind the AI boom (or bubble)’, has outlined some very instructive examples of this.

Take Nvidia, for example, the manufacturer of specialised processors used to train and run large artificial intelligence systems. These processors are in high demand, are quite expensive and are an essential component of the massive data centres being built all over the world. Nvidia makes money by selling them, but it also invests in the companies that buy them.

Among these is OpenAI, the company that develops ChatGPT. Nvidia is a financial backer of OpenAI, which requires enormous computing power and therefore Nvidia processors. The manufacturer provides financial support to one of its own customers, who can use the resources received to purchase the financier’s products. Nvidia has also made financial commitments relating to the development of infrastructure for artificial intelligence, thereby helping to further sustain demand for its products.

Of course, investments and supplies are separate transactions, and their amounts must be considered in accordance with the agreements actually concluded. However, the economic relationship that is created is quite straightforward: part of the money used to finance clients’ growth comes from the very companies that profit from their expenditure.

There is also an even more curious example: Firmus, an Australian company that builds data centres. Nvidia has participated in its funding and provides it with the technology needed to build the infrastructure. Firmus has also entered into an agreement to supply computing capacity to Nvidia itself, with a potential value of up to 30 billion dollars over six years. The same company can therefore act simultaneously as an investor, supplier and customer. In turn, OpenAI has been announced as a key client of some Firmus centres in Malaysia.

Anyone looking at these companies’ results sees investments running into billions, new orders and contracts of staggering proportions. These are all genuine economic activities, but the way they are intertwined makes it difficult to determine how much of the overall growth depends on demand from outside the system – demand capable of generating sufficient revenue to sustain it over time.

The issue has also been studied by the Bank for International Settlements (BIS), an international financial institution based in Basel, founded in 1930, which promotes cooperation between central banks and studies risks to the stability of the global financial system. In a report published on 1 October, the BIS examined investments between artificial intelligence firms over the period 2021–2025. The findings are interesting: 55.2 per cent of the value of investment received by companies in the sector came from other companies in the same sector. Furthermore, when considering investments made by one artificial intelligence company in another, 46.4 per cent of their value related to transactions in which there was also a commercial supply relationship between the investor and the recipient.

These percentages do not prove that the money was used entirely to purchase the funder’s products, nor do they allow us to calculate what proportion of the companies’ revenues derives from circular financing. However, they do illustrate just how important cross-shareholdings and financial relationships have become between companies that are simultaneously each other’s customers and suppliers.

To understand why this might be a problem, let’s imagine that the forecasts regarding the future use of artificial intelligence turn out to be overly optimistic. The companies that built the data centres would be left with costly infrastructure and insufficient revenue. They would cut back on purchases of new processors, harming suppliers. The latter would see their sales decline whilst, at the same time, the value of their investments in their customers would diminish. The worsening outlook could make it more difficult to secure new funding, leading to further reductions in investment.

In a system of this kind, a company’s difficulties can therefore quickly spread to others, precisely through those links that had initially fuelled its growth. The Reserve Bank of Australia, too, in its October financial stability report, drew attention to the growing opacity of circular financing and the risks arising from expectations of particularly rapid profit growth.

There is a rather instructive historical precedent. In the 1990s, during the expansion of the Internet, some telecommunications equipment manufacturers subsidised their customers to encourage them to purchase their technologies. Expectations were sky-high, and indeed the Internet would go on to transform the world. The problem was that much of the infrastructure was built before there was sufficient demand to make it economically viable. When the telecoms bubble burst, several companies went bankrupt and many investors lost huge sums of money.

The internet naturally continued to develop, drawing in part on the infrastructure built during the speculative phase. The validity of the technology had not prevented expectations regarding the pace of its adoption from leading to excessive investment.

In the case of artificial intelligence, there is also an additional challenge. Specialised processors become technologically obsolete within a few years, due to the rapid development of new generations of chips. Those who have invested billions in purchasing them must therefore recoup the cost fairly quickly. If the expected revenues are slow to materialise, the time available to recoup the investment is reduced.

Ultimately, everything depends on how quickly artificial intelligence manages to generate a market profitable enough to sustain the immense infrastructure currently being built to serve it. The services are already used by millions of people and a great many businesses, but this alone does not allow us to determine whether future revenues will be commensurate with current investments.

The scientific and technological success of artificial intelligence can therefore coexist with the financial overvaluation of the companies that develop it. A technology can profoundly change the economy and, at the same time, cause colossal losses to those who have bet too early on its profitability.

Circular financing makes this venture particularly delicate, as it allows a portion of the demand one hopes to meet tomorrow to be anticipated today. As long as the money keeps coming in, everyone can show encouraging results. Ultimately, however, someone must pay for the services provided by the data centres, using resources earned outside the funding circuit that built them. And it is the scale of that demand – far more than the billion-euro investments announced by companies – that will determine the sector’s economic stability.

There is, however, a further factor that risks fundamentally altering the terms of the issue: the growing pressure for governments to shoulder an ever-increasing share of the demand for artificial intelligence. If the private market does not grow quickly enough to justify the investments already made, public authorities may become particularly valuable customers, as they have budgets funded by general taxation and can commit to long-term procurement programmes. Part of the risk borne by private investors would thus end up being transferred, at least indirectly, to taxpayers.

Consider, for example, the persistent push for the widespread adoption of artificial intelligence in schools. In January 2026, OpenAI launched the ‘Education for Countries’ programme, through which it proposes to governments and university systems the integration of its tools into teaching, teacher training and student activities. Among the first participants are Estonia, Greece and the Conference of Italian University Rectors. In Estonia, ChatGPT Edu has already been rolled out nationwide, reaching tens of thousands of pupils and teachers. In the United States, OpenAI has announced agreements with over a hundred educational organisations, offering free access and training to more than 300,000 educators and staff members. The initial free access encourages the widespread adoption of these tools and may create the conditions whereby, once integrated into daily activities, it becomes difficult to do without them or replace them. The result could be the creation of a stable demand for artificial intelligence services, supported in future by public budgets, regardless of whether their use is shown to actually improve learning sufficiently to justify the costs.

The same process applies to the public sector. In September 2026, OpenAI announced an agreement with the US government providing free access to licences and a 50 per cent discount on usage charges for a wide range of federal and local government bodies, with the agreement set to run until the end of 2028. Here too, the initial cost savings may encourage the adoption of tools destined to become standard components of administrative work, creating a technological dependency that could result in recurring expenditure in subsequent years. In the healthcare sector, the National Health Service announced in June 2026 that it would roll out Microsoft Copilot to over half a million staff, based on preliminary trials which reportedly demonstrated time savings in administrative tasks. These are potential benefits that warrant independent and ongoing scrutiny, particularly when the purchase is funded by public money and large-scale roll-out precedes a comprehensive assessment of the effects.

In addition to these acquisitions, there is a call for public support for the infrastructure required for artificial intelligence, through tax incentives, favourable conditions for the construction of data centres, and investment in the electricity grids needed to power them. Here too, part of the costs associated with the sector’s expansion can be passed on to the public, whilst the revenue from the infrastructure and services remains in the hands of the companies that manage them.

There is no need to imagine a coordinated plan amongst companies to understand the economic benefits of this strategy. Every company has an interest in expanding its customer base and securing favourable conditions for growth. However, when an entire sector has invested enormous sums based on forecasts of future demand that has yet to be proven, promoting public adoption also serves as a form of insurance against those forecasts failing to materialise. Presenting the use of artificial intelligence as a necessity for the modernisation of education, healthcare and public administration can help transform a technological choice – which ought to be justified by results – into an expenditure deemed inevitable.

The ultimate risk, therefore, is that, having mutually financed their own expansion and achieved colossal market valuations as a result, artificial intelligence firms will succeed in persuading governments to use public funds to guarantee a portion of the demand needed to sustain them. If the forecasts prove correct, private investors will have made extraordinary profits. If they prove to be over-optimistic, part of the losses could be absorbed through public procurement, incentives and infrastructure funded by taxpayers. This would be a particularly advantageous distribution of risk for those who took the gamble, as it would leave the private sector with the lion’s share of the benefits of success whilst placing an increasing share of the costs of any failure on the public purse. The question we should be asking governments, before any new programme involving the mandatory or widespread adoption of artificial intelligence, is therefore very simple: are we purchasing a service whose usefulness we have demonstrated, or are we helping to foot the bill for private investments that have yet to prove their sustainability?

To understand why this discovery is significant, it is important to bear in mind that multiple sclerosis is a chronic inflammatory disease of the central nervous system, in which an abnormal immune response primarily attacks myelin, the lipid-rich sheath that surrounds axons – the extensions through which neurons transmit electrical signals. Myelin enables nerve impulses to travel rapidly and efficiently; consequently, its destruction causes disturbances in nerve conduction, which can manifest as vision problems, changes in sensation, muscle weakness and difficulty coordinating movements. When the inflammatory process persists, the axon itself may also be damaged, with potentially permanent consequences.

The therapies currently available target different components of the immune response, and some have proved extraordinarily effective in preventing new episodes of inflammation. However, once significant neurological damage has occurred, suppressing pathological immune activity may not be sufficient to restore lost functions. This distinction explains why a molecule capable of improving the condition of animals that are already symptomatic is of particular interest, especially when the clinical effect is accompanied by evidence of better preservation of myelin and axons.

However, the scientific journey that led to the new drug began far removed from multiple sclerosis, in studies on age-related macular degeneration, a retinal condition that can severely impair central vision. In 2014, Ambati and his colleagues published a study in Science demonstrating that certain drugs already used to treat HIV possessed a previously unexpected property. These substances, known as nucleoside reverse transcriptase inhibitors, or NRTIs, had been designed to interfere with an enzyme essential for retrovirus replication. Yet, in experiments conducted by the American team, they were found to be capable of preventing the activation of a specific molecular system responsible for certain inflammatory responses, independently of their antiviral activity.

The system in question is called the inflammasome. It is a protein complex that certain cells assemble when they recognise danger signals, such as those associated with cellular damage or infections. Once activated, the inflammasome promotes the production of potent inflammatory mediators and can trigger a form of cell death known as pyroptosis, which helps to amplify the immune response. One of its main components is the NLRP3 protein, which is involved in numerous conditions characterised by persistent inflammation.

In the 2014 experiments, NRTIs were able to block the activation of the NLRP3 inflammasome induced by specific RNA molecules, by interfering with the response associated with a cell receptor called P2X7. The result suggested a much broader therapeutic potential than that for which these drugs had originally been developed: molecules designed to block a viral enzyme could be used to modulate certain inflammatory processes in the body.

This observation gave rise to the idea of chemically modifying the NRTIs to preserve their anti-inflammatory activity, whilst seeking to reduce the side effects associated with some of the original molecules. The resulting family of compounds was named kamuvudine. Among these, kamuvudine-9, abbreviated to K-9, was selected to investigate whether blocking the inflammasome might also have therapeutic effects in a neuroinflammatory disease such as multiple sclerosis.

To test this hypothesis, the researchers used an experimental model known as experimental autoimmune encephalomyelitis, or EAE, in which an immune response against components of myelin is induced in mice. The animals thus develop an inflammatory disease that replicates certain key aspects of human multiple sclerosis, including motor deficits and visual impairment. The model naturally has its limitations, as the human disease is far more complex and can progress through mechanisms that are not fully replicated in the animal. It remains, however, one of the most widely used tools for studying the consequences of autoimmune inflammation in the nervous system.

The most interesting aspect of the new study concerns the point at which treatment began. The researchers also administered K-9 to animals in which neurological symptoms had already appeared, continuing the treatment for around two weeks. In the treated mice, disease progression was halted and a recovery in motor and visual abilities was observed; this recovery was not achieved to the same extent in the animals in the control groups, including the group receiving conventional treatment for multiple sclerosis.

This observation alone could have been due to a reduction in inflammation, allowing the nerve structures that were still intact to resume functioning. For this reason, the authors also looked for signs of nerve tissue protection by examining the animals’ spinal cords. In mice treated with K-9, they found better preservation of myelin and axons, accompanied by a particularly significant finding: the treatment prevented an increase in the blood concentration of the neurofilament light chain, denoted by the abbreviation NfL. Neurofilaments are structural proteins found mainly in axons and are released when these structures are damaged. Their concentration in the blood can therefore provide an indirect measure of the activity of neurodegenerative processes.

The correlation between improved neurological function and these indicators of nerve tissue protection makes the result more significant than a mere reduction in symptoms. It remains to be determined how much of the observed recovery stems from the survival of previously compromised nerve structures and how much from any repair processes. In particular, the study does not demonstrate that K-9 is capable of regenerating destroyed axons or of rebuilding myelin that has already been permanently lost. The possibility of regaining neurological function after the onset of a deficit is, however, an important finding, precisely because it suggests that some of the functional damage may be reversible when the appropriate inflammatory mechanisms are targeted.

Molecular experiments have also yielded a result worthy of attention. K-9 interferes with the interaction between NLRP3 and a protein called NEK7, which is necessary for inflammasome activation under certain conditions. It also hinders the association of NLRP3 with NLRC4, another protein involved in cellular danger recognition systems. Accordingly, the authors propose that the molecule acts on two pathways of inflammasome activation, thereby limiting an inflammatory response capable of contributing to the degeneration of nerve tissue.

This mechanism allows the results to be interpreted within a broader context. In multiple sclerosis, in fact, the adaptive immune system – which is responsible for the specific recognition of antigens via lymphocytes – plays a central role in triggering the autoimmune response. However, a significant part of the damage to the nervous system also depends on the activity of innate immune cells, including microglial cells, which reside in the brain and spinal cord and are involved in the response to damage. The persistent activation of these systems may contribute to the persistence of inflammation and axonal damage, even when some of the immune processes that initially triggered the disease are controlled by medication.

The potential to influence these mechanisms is particularly significant in progressive forms of multiple sclerosis, where disability can continue to worsen even in the absence of obvious clinical relapses. It would be premature to conclude that K-9 can alter the course of these forms of the disease, as the model used in the study reproduces only part of their complexity. The research does, however, point to a possibility that warrants further investigation: targeting the inflammatory processes that contribute directly to neurological damage could complement therapeutic strategies aimed at the autoimmune response.

The authors sought to bolster this interpretation by also using data from human patients. They examined three distinct populations using large US health and insurance databases, covering a total of over three million people. The analysis included individuals treated with NRTIs for HIV infection, people taking these drugs to prevent infection, and patients undergoing antiviral therapy for hepatitis B. In these populations, exposure to NRTIs was associated with a lower incidence of multiple sclerosis compared with the control groups. Among individuals who had already been diagnosed with multiple sclerosis, taking NRTIs was also associated with a lower frequency of subsequent relapses.

This is a striking finding, as it stems from human observations independent of experiments conducted on mice and concerns a family of molecules that, like K-9, have the ability to interfere with certain inflammatory mechanisms. It must be borne in mind, however, that these analyses are observational. People who take antivirals differ from those who do not in terms of clinical conditions, exposure to infection and patterns of healthcare, all of which are factors that can influence the risk of diagnosis and the course of the disease. Even a statistically robust association, after appropriate adjustments, cannot replace experimental proof of a treatment’s efficacy in patients.

There is also an important biological distinction. The NRTIs used against HIV and hepatitis B interfere with enzymes necessary for viral replication, whilst K-9 has been developed specifically to separate its anti-inflammatory activity from its antiviral activity. The epidemiological data collected on traditional NRTIs therefore support the plausibility of the proposed mechanism, without proving that the new molecule possesses the same efficacy in humans as that observed in animals.

The link between viral infections and multiple sclerosis adds a further element of interest. In 2022, a study published in Science, based on the monitoring of millions of US military personnel, showed that infection with the Epstein-Barr virus almost invariably preceded the onset of the disease and was associated with an approximately 32-fold increase in the risk of developing it. Other research has identified possible mechanisms through which the immune response to the virus might also recognise proteins in the nervous system. This phenomenon, known as molecular mimicry, occurs when different antigens share sufficient similarities to trigger cross-reactive immune responses.

These findings have reinforced the hypothesis that Epstein-Barr virus infection represents a key step in the development of multiple sclerosis, although further conditions are required for the disease to manifest itself. The new study on kamuvudine, on the other hand, focuses on the processes that contribute to neurological damage once the pathological response has taken hold. The two lines of research are compatible and could contribute to a more comprehensive understanding of the disease, in which the factors that promote its onset and the mechanisms responsible for the subsequent nerve degeneration play different roles.

However, there are still crucial steps to be taken before considering a therapeutic application. It will be necessary to verify the safety of K-9 when administered over prolonged periods and to establish whether the concentrations achievable in the body are sufficient to modify inflammatory activity in the human nervous system. The molecule has already entered clinical trials for certain eye conditions, including diabetic macular oedema, and this may provide useful information on its behaviour in humans. However, its efficacy in multiple sclerosis remains to be demonstrated through controlled clinical trials capable of measuring both disease activity and any recovery of neurological function.

Meanwhile, the story of the nucleoside analogues demonstrates just how fruitful it can be to study the biological effects of drugs beyond the target for which they were designed. A compound developed to inhibit viral replication has revealed a second mode of action, from which it has been possible to derive a new family of molecules with different characteristics. The journey from AIDS research to age-related macular degeneration and finally to multiple sclerosis shows how an understanding of molecular mechanisms can suggest applications that nobody would have imagined when the original substances were first introduced.

If the results obtained in animals are confirmed in humans, the most significant benefit could be for precisely those patients for whom preventing new lesions is no longer sufficient. For those who have lost the ability to walk unaided, or have seen their vision progressively deteriorate, even partial recovery can represent a huge improvement in quality of life. Proof that this is possible with K-9 is yet to be provided, and it will be clinical trials that determine whether the promise observed in mice can be translated into a treatment. Today’s result, however, justifies a question that until a few years ago would have seemed particularly ambitious: to what extent can we restore the functions of the nervous system when we learn to control the causes of its deterioration?

As the model is not publicly accessible, no one outside OpenAI is able to verify these claims, which is why, on 29 September, the advisory group on mathematics and artificial intelligence hosted by the Institute for Advanced Study in Princeton had asked the laboratories to accompany each result with the prompts from which it originated and the associated computational cost; OpenAI responded to this request by providing only an average of the computational effort required and by releasing ten abbreviated summaries of the model’s reasoning, without publishing any prompts. The error, announced on 7 October in the repository’s change log, consists of an incorrect sign within a manuscript on the algebraicity of Weil classes for certain abelian varieties of dimension eight, which invalidates a cancellation argument and, in turn, undermines the construction upon which two other papers were based, including the one that established the rational Hodge conjecture for products of K3 surfaces. The three manuscripts belonged to the family of results linked to the Hodge conjecture—the most closely watched in the entire catalogue—and the flawed one, dated 18 September, was among the oldest in the collection.

On the same day, OpenAI corrected a further fourteen manuscripts, rectifying proofs and clarifying statements and hypotheses, and updated the bibliographical references of a further thirteen, meaning the catalogue has been reduced from 722 to 719 titles, whilst the withdrawn versions remain available for consultation alongside a note describing the omission. A spokesperson attributed the discovery to an internal audit, without specifying who had identified the error, and assured that the review process, as is the case for any research manuscript, will continue until those papers for which no correction can be found are withdrawn. OpenAI usually responds to those who raise concerns by referring to Lean, an open-source language in which a proof is rewritten in rigorous form so that a programme can check every step; at the time of publication, however, only 162 of the 722 manuscripts had their main result formalised, and with the additions of 7 October, this figure rose to 300 main results out of 719 – just over 42 per cent – a proportion that has remained unchanged since then, given that the change log still contains only that single entry and the README repeats the same percentage. The company acknowledges that some of the non-formalised results may contain errors and explains that it published them without waiting for full formalisation on the recommendation of the advisory group itself, although a comparison with the repository index shows that almost everything that has been withdrawn or substantially revised was outside the part verified by Lean.

The fact that even the verified part requires caution is demonstrated by the precedent of 8 September, when OpenAI announced a 166-page proof on the formation of singularities in finite time in fluid equations, produced by around ten thousand agents in 88 hours and formalised in Lean in a further 17 hours; however, this concerns equations subjected to an external force, whereas the Millennium Problem requires the case without an external force to be addressed, which is why the Clay Institute does not consider it solved. On 6 October, the same day the 722 manuscripts appeared, Alexander Bastounis, Fabian Circelli and Anders C. Hansen published a preprint on arxiv entitled ‘Navier-Stokes lost in translation’, in which they argue that the proof formalised in Lean does not correspond to the proof written in natural language, which allegedly asserts more than the code verifies, and they cite other examples of erroneous translations produced by artificial intelligence systems alongside this case, going so far as to argue, on a theoretical level, that ensuring the fidelity of a formalisation is an intrinsically more difficult problem than the Turing halting problem.

As this is a preprint, which OpenAI does not appear to have replicated and whose abstract does not specify any particular files or versions, its conclusions must be scrutinised, bearing in mind that a formal proof following a different path from the one written leaves the result intact as long as the formalised statement coincides with the theorem one sought to prove, and that the problem only arises when the statement itself diverges. The issue, however, concerns all 719 manuscripts, since Lean guarantees that the code demonstrates what is written in the code, whilst establishing that what is written there coincides with the theorem of interest to mathematicians remains a human task that allows for no shortcuts; thus, that 42 per cent figure merely measures the proportion of the catalogue that has been scrutinised by the machine, without indicating how much of it is actually correct.

The retraction comes at the end of a month of tensions, which began with the dispute over priority that had accompanied the announcement regarding the Navier-Stokes equations, when it emerged that two mathematicians, Alpöge and Buckmaster, were working on the same problem, and continued on 11 September with the statement in which twenty-five Fields Medal winners, including Terence Tao, criticised the use of mathematical problems as a testing ground for the artificial intelligence industry. Ursula Martin, a mathematician at Oxford, compared the bulk publication of the 722 manuscripts to the submission of a messy first draft that the community would then have to sort out, and this comparison is borne out by the figures, given that at OpenAI each manuscript cost an average of three hours of computation, whilst verifying it requires weeks or months of specialist work that falls on researchers who did not ask for it. The company promises to fund workshops and conferences to help mathematicians understand the results and assures that it is working on releasing the model, whilst the advisory group has described the publication as the beginning of human understanding rather than its conclusion, suggesting that the bulk of the work has yet to begin and that it will fall to others to carry it out.

A sign error is perhaps the most human of all errors, one that befalls students sitting their first exam just as much as Fields Medal winners, and what this incident calls into question is rather the distribution of effort, because mathematics rests on a pact whereby whoever signs off on a proof understands and defends it, whilst the community undertakes to verify it; that pact falters when the author is a model that no one can question and the reviewers are mathematicians who had not asked for any of this. Lean was supposed to offer a way out by entrusting the verification to software, but the Navier–Stokes affair serves as a reminder that software only verifies what is submitted to it, and as long as OpenAI continues to produce theorems faster than the world can read them, the word ‘proven’ will have to be taken with a pinch of salt, even when certified by a computer.

Science provides a particularly clear example. In 2025, according to the estimate cited by Helen Pearson in Aeon, around seven million scientific articles were published – over 19,000 a day. Each study may contribute a useful finding; when presented in isolation, however, it can create a misleading impression. Newspapers report on the latest research; readers are confronted with headlines that appear to contradict one another, and the bigger picture remains out of focus.

This happened with paracetamol during pregnancy. In September 2025, the US regulatory agency, the FDA, issued a warning based on a possible link between taking the medicine and autism in children. The news caused a great stir. A subsequent review of the body of evidence concluded that the available data did not clearly show a link to autism or ADHD. In the meantime, the alarm had reached many women, while the comprehensive assessment required the less high-profile work of reviewing the literature and evaluating the quality of the studies.

In cases such as this, the necessary knowledge is not limited to a single publication. It takes shape by comparing results, distinguishing the most robust studies and assessing what they indicate as a whole. A systematic review brings together relevant studies according to stated criteria and assesses their reliability; where the data are comparable, a meta-analysis combines the results. These methods were developed precisely to make sense of a body of literature which, when viewed as the sum of individual articles, may appear contradictory.

The history of the recommended sleeping position for newborns illustrates the cost of delayed action. For decades, it was recommended that babies be placed on their tums. A subsequent review revealed that the risks associated with that position were already apparent in studies conducted around 1970 – some twenty years before the advice changed widely. Researchers estimated that a timely synthesis of the evidence, followed by appropriate action, could have prevented at least fifty thousand deaths in Europe, Australasia and the United States. The estimate is counterfactual, and for this very reason should be interpreted with caution; nevertheless, the case demonstrates that the accumulation of findings is not sufficient to produce the kind of knowledge capable of guiding decisions.

Science has developed tools to carry out this work. Gene Glass’s meta-analysis in the 1970s helped move beyond the rudimentary counting of favourable and unfavourable studies, which assigned equal weight to research of varying sizes and with differing effects. The establishment of the Cochrane Collaboration in 1993 brought the concept of systematic review to the forefront of medicine. Yet the scientific system continues to prioritise the publication of new findings above all else. Synthesis takes years, does not always produce a catchy headline and struggles to compete for attention with the latest study.

Digital growth also turns this challenge into a question of power. To access scientific literature, we rely on search engines, platforms and artificial intelligence systems that rank pages and decide which results to show us. Their selection process is inevitable; the way in which it is carried out often remains opaque. The few large companies that control these tools thus become gatekeepers of scientific information: they determine which content comes to the fore, which is summarised and which remains buried on subsequent pages. This selection process influences what the public can learn and what evidence informs decision-making.

A response generated by artificial intelligence makes the filtering process even less transparent. The user receives a pre-organised conclusion, without always knowing which studies were consulted, which were excluded, or how reliable the sources considered are. Some tools can help identify and classify articles, speeding up parts of the work; however, their ability to produce reliable reviews must be demonstrated, and an automated summary may provide an incomplete picture of the literature. Efficiency in locating publications is not the same as the ability to evaluate them.

The public issue, then, concerns who controls attention and according to what criteria. If the filtering is entrusted to proprietary systems, whose criteria cannot be verified, the formal availability of articles coexists with access governed by private decisions. We need transparent search tools and up-to-date scientific summaries that are accessible and verifiable; it is also essential to make the sources and the process leading to a conclusion transparent. Scientific knowledge is built when results are organised and subjected to scrutiny. In an environment saturated with information, those who decide what deserves attention wield decisive power over what society is able to know.

Although this discovery has nothing to do with the psychedelic properties of psilocybin – and therefore with the hallucinatory effects experienced when taking hallucinogenic mushrooms – it nevertheless speaks to the rehabilitation that psychoactive substances have been undergoing for the past twenty years or so. That is, what is generally referred to as the ‘psychedelic renaissance’. It is no coincidence that, in Australia, since 2023, specifically authorised psychiatrists have been able to prescribe psilocybin to people diagnosed with treatment-resistant depression. But what happens to the brain when psilocybin is taken? This was the question posed in 2012 by a group of researchers led by Carhart-Harris, a professor at Imperial College London. In this study, the researchers used functional magnetic resonance imaging to observe how the brain responded to psilocybin. The results showed that levels of brain activation decreased significantly, even though it had previously been commonly assumed that psychedelic substances worked by increasing neural activity.

The fact that activity levels were reduced in the areas of the brain responsible for emotional regulation and decision-making, such as the anterior cingulate cortex, has led to the hypothesis that psilocybin acts as a sort of ‘relief valve’, as a means of deactivating the personal frameworks used to experience and interpret the world, ‘allowing for a state of uninhibited cognition’. It is as if to say: ‘Veil of Maya, adieu!’ Indeed, the authors emphasise that people suffering from depression exhibit excessive activity in the ventromedial prefrontal cortex, which manifests as rigidity and pessimism, as well as a tendency to slip into negative moods. According to the findings of their research, it is precisely this excess activity that psilocybin appears to dampen. While not all results are conclusive, studies conducted during these years of the ‘psychedelic renaissance’ appear encouraging.

Above is the live press conference from Stockholm.

Last year, the prize was awarded to Susumu Kitagawa, Richard Robson and Omar Yaghi for the development of metal-organic frameworks: porous materials capable of capturing gases, storing substances and separating molecules, with applications ranging from water collection in the desert to CO₂ capture.

The Nobel Prize in Chemistry is the third to be announced this week. Yesterday, the prize for Physics was awarded to the Belgian Francis Halzen, for his role in the creation of IceCube, the observatory that uses Antarctic ice to detect neutrinos from deep space. The other announcements will follow in the coming days: Literature, Peace and, on Monday, Economics.

Physicist Francis Halzen laid the foundations for a completely new branch of astronomy. His idea was to use glacial ice at the South Pole to capture the elusive particles known as neutrinos. When a neutrino collides with an atomic nucleus, it produces a flash of light that can be detected by sensors within the transparent glacial ice. The ice at the South Pole offers numerous advantages, as it is free from various types of interference and the area is geologically stable, with no earthquakes. Halzen and his idea soon gained the support of other researchers and, just a few years later, preliminary tests were carried out on the sensors in the ice. The result was the IceCube Neutrino Observatory, completed in 2011, capable of detecting neutrinos from the distant cosmos and consisting of one cubic kilometre of ice fitted with light sensors. Thanks to IceCube, researchers can detect neutrinos originating from extremely high-energy processes in the distant universe.

Researchers soon detected the first high-energy neutrinos and, a few years later, were able to publish their discovery of neutrinos that must originate well beyond our solar system. The search for neutrino sources in the universe could thus begin in earnest. The neutrino interactions continuously detected by IceCube will provide researchers with new insights into the violent environments in which high-energy neutrinos can form, and may even reveal previously unknown cosmic phenomena.

Last year’s Nobel Prize winners in Physics were John Clarke, a professor at the University of California, Berkeley; Michel H. Devoret, a professor of applied physics at Yale University and the University of California, Santa Barbara; and John M. Martinis, “for the discovery of the macroscopic quantum tunnelling effect and the quantisation of energy in an electrical circuit”. Meanwhile, for the 2026 awards, the winners of the Nobel Prize in Medicine – Karl Deisseroth, Peter Hegemann and Georg Nagel – were announced yesterday, whilst tomorrow will see the announcement of the prize for Chemistry, followed by the prize for Literature on Thursday the 8th. Finally, the Peace Prize will be announced on Friday the 9th.