What appeared to be an isolated incident was followed by a series of events that raised the alert level four months later. In early July, an actor suspected of having links to China carried out twelve waves of attacks against Taiwan’s government infrastructure. Starting from a single portal, the agents mapped twenty-one connected systems, compromised 85 accounts and extracted over 2,500 personnel files. The attackers identified a flaw in the signature validation of the authentication service and exploited it to leave a backdoor open. The tools used were open-source, including OpenClaw itself. Taipei confirmed the intrusion in mid-August without naming a culprit.

There is also a precedent for this particular case dating back to 2025, when Anthropic described a campaign by a Chinese group in which AI carried out between 80 and 90 per cent of the tactical operations on its own, whilst humans selected the targets. Security firms documenting today’s cases describe it as a machine that makes decisions autonomously. Furthermore, in a session observed by Unit 42, a model attempted to exploit a specific vulnerability: when the attempt failed, it judged the target to be of little value and sought a better one from among the most popular exploits on public repositories. No human had asked it to change its target.

In short, cybersecurity may certainly benefit from AI in the future, but at present it is primarily hackers – both lone wolves and state-sponsored actors – who appear to have benefited most from experimenting with these agent-based tools. Indeed, the safeguards of AI systems, where they exist, are easily breached. In TrendAI’s half-yearly report, an agent working for a pro-Chinese group was taken in by a simple lie: that the operation was an authorised penetration test. From that moment on, he carried out extensive reconnaissance and a blanket harvest of credentials within the target network.

Across the vast landscape of AI agents, even seemingly marginal countries have stories to tell, and there are also those who prefer not to deal with a supplier but to work directly in-house. The North Korean group Kimsuky had installed its models locally, where no one can read the prompts, and in September, researchers at Genians found traces of an open-source programming agent in the metadata of PDFs that were concealing its malware.

The incident and the attack must be viewed together, as they are the same phenomenon seen from two different angles. Last May, an attacker sent an NFT capable of activating elevated permissions to the Bankr wallet – an assistant that carries out transactions on command – and managed to extort around 175,000 dollars. In the age of AI, every agent essentially acts as if it had a text-based remote control: it can be activated with the right phrase, and this phrase may be written by the legitimate owner or by an intruder with a credible cover story.

European regulation, which does its best but is obviously perpetually lagging behind a technology that is advancing at an unprecedented pace, focuses primarily on the provider of the model. However, these incidents take place elsewhere, between the agent and the person speaking to them, in a space that no regulation assigns to anyone.

An agent can therefore become rogue, or malicious, if the chain of command breaks down. But whilst it used to require highly advanced and sophisticated tools to gain access to the privileges of that chain, today all it takes is a single intrusion and a single phrase at the right moment to trigger increasingly serious consequences – all the more so as we grant greater power and control to these AI-powered agents.

This refers to the Maven Smart System, used in the capture of Maduro and in the attacks in Iran. And amongst the fifty-one pages of the court order, including the dissenting opinion of Judge Karen Henderson, the precision operation to arrest the Venezuelan president is mentioned as the main point of contention between Anthropic and the DoD. According to executives at the company that owns the Claude LLM, the Pentagon allegedly breached Anthropic’s own guidelines against using its models for the ‘mass surveillance of American citizens, even for lawful purposes’ and for the implementation of autonomous weapon systems. For their part, administration officials have accused Anthropic of triggering the LLM’s internal safety safeguards, thereby jeopardising the success of the operation and the safety of the personnel involved. Furthermore, it appears that Anthropic had also refused similar requests from health authorities, in this instance the Centres for Disease Control and Prevention (CDC).

The Maven interface represents one of the highest-calibre elements of technological warfare and intelligence and is a flagship project for the Pentagon, which has allocated $2.3 billion for Maven alone in 2027, having, as of now, replaced Anthropic with OpenAI. The brainchild of the tenacious and visionary commitment of Marine Colonel Drew Cukor, who began working on it in 2017 alongside the major tech giants, Maven faced a revolt by Google employees who were initially involved. Following Google’s withdrawal in 2018, Palantir stepped in and established itself as the main provider of the digital infrastructure. Katrina Manson, in her book ‘Project Maven’, describes the project as the dawn of war with artificial intelligence, whilst Palantir’s CTO Shyam Sankar, in ‘Mobilize’, details Cukor’s operational and bureaucratic odyssey as he sought to drastically cut red tape. Since March 2025, the interface has also been adopted by NATO, providing yet another reason to gain a deeper understanding of it. It functions as an infrastructure for integrating data from any available source – drones, cameras, databases and intelligence reports – for the purposes of conflict zone management and target identification, strategic planning and attack planning using artificial intelligence and machine learning, and the refinement of decision-making and intelligence processes. From a human perspective, Maven’s primary aims are to accelerate and optimise the analysis of massive volumes of data, overcoming the challenges of harmonising disparate sources and information resources. It incorporates large language models (LLMs), primarily—though not exclusively—Anthropic’s Claude. Unlike conventional computer systems, LLMs enable far more refined and comprehensive analytical insights. In ‘The Technological Republic’, Alex Karp denounced the collective amnesia of Google’s engineers, who, by refusing to work on Maven, have contradicted the history of Silicon Valley. Karp made similar remarks in a TV interview regarding Anthropic. He stated that this is a form of instrumental pacifism that risks jeopardising the security of the West.

Just a few days ago, the public debate was dominated by catastrophic scenarios following statements by researchers at Anthropic – with internal warnings that AI could lead to human extinction within the decade – which were followed by official statements from CEO Dario Amodei, who called for a controlled slowdown in the sector, and public support from figures such as Elon Musk. Yet, with disarming nonchalance, the very same premises are now being turned on their head to foreshadow the imminent eradication of almost every human disease. Amidst this swing between catastrophism and biological utopia lies the latest announcement from Anthropic, the company developing the Claude model.

The scientific fact, stripped of all propaganda, is undoubtedly intriguing. Set loose on a vast genomic database with just a single guiding prompt, Claude co-ordinated around 950 agents in parallel for 21 consecutive hours. It sifted through hundreds of thousands of DNA sequences, narrowed down the field with the tenacity of a bloodhound, and finally identified something that had eluded human biologists: an enzymatic system based on a reverse transcriptase associated with a series of uniformly spaced repetitive sequences. The researchers have named it ART (array-associated reverse transcriptases). A molecular arrangement that closely resembles the structure of CRISPR, the bacterial immune system that has revolutionised gene editing over the last decade.

In biological terms, the discovery is genuine. Even Feng Zhang, a professor at MIT and one of the pioneers of molecular scissors, acknowledged this, describing the result as “genuinely exciting”. And Amodei himself, with self-deprecating humour measured down to the millimetre, commented that the discovery of this molecular machine in bacteriophages represents “a piece of work I would have been proud of as a PhD student”. The machine read the literature, sifted through the data, spotted the anomaly and even suggested the bench-top experiments that real-life scientists then carried out in the laboratory to confirm that the array is indeed expressed in short RNA sequences.

And yet, reading the pre-print released by Anthropic with a touch of scepticism, it is difficult to separate the scientific revelation from the public relations moment. The publication comes hot on the heels of yet another round of controversy and reports on the ‘existential risks’ associated with next-generation models. Anthropic’s move has a distinctly defensive flavour: what better way to allay fears of a destructive AI than to show the world that your model is laying the foundations for the medicine of the future?

What is needed, in fact, is a touch of secular clarity. As the authors of the paper themselves admit with a modicum of modesty, the precise biological function of ART remains unknown. We do not know whether it really functions as an editing tool, nor whether it will ever have any concrete biotechnological utility or be applicable to humans. To turn the identification of a genomic pattern into the implicit promise of a universal cure is to confuse the first stage of a Grand Prix with the finish line. Biomedical progress is an gruelling marathon stretching from basic biology to clinical trials, via the complex process of delivering medicines to patients; a journey where failures in the test tube number in the thousands.

This in no way detracts from the value of the method. Seeing an agent-based architecture process 210 million tokens to carry out, in less than a day, a sifting task that would have taken a team of researchers months is confirmation that the trajectory of AI applied to the life sciences is not merely a marketing ploy. Amodei often refers to his essay "Machines of Loving Grace" and the exponential trajectory of models which, having stumbled over secondary school equations in 2023, have now reached the point where they are tackling open problems in mathematics. Biology could follow a similar trajectory.

But in the meantime, amidst press releases and pre-prints – whether the brainchild of the model itself or the PR department’s strategy – we would do well to keep a cool head. Artificial intelligence will not wipe us out by 2030, but nor will it cure all the world’s diseases within the same timeframe. In between lies real science: made up of brilliant algorithmic insights, human laboratories tasked with verifying the data and, above all, a public that must learn to distinguish between the advancement of knowledge and a sophisticated public relations campaign.

As the election approaches, the issue has been addressed informally by the leaders: Giorgia Meloni and Elly Schlein have endorsed Il Foglio’s call for a moratorium on deepfakes targeting political opponents during the election campaign. The Butti bill also stipulates that deepfakes must be clearly labelled, otherwise it constitutes a criminal offence. The question, however, is whether this will be enough. Centinaio wonders about this, whilst Ascani is sceptical: “The difference between a law and a gentlemen’s agreement,” says the Deputy Speaker of the Chamber of Deputies, “is that the law is binding: if it is breached, it provides the means to enforce it. We have committed to the moratorium, but I do not know if it will be enough: if this tool is left available, there will be those who use it. Not so much the political parties, but those ecosystems of pages and accounts that are not directly attributable to any political force. “We have a problem ahead of the next election campaign, but we’re heading into it empty-handed,” concludes the Deputy Speaker of the Chamber of Deputies. Centinaio agrees: “We are in favour of the free dissemination of even the worst things in the world, but showing content distorted by AI in a covert manner influences voters and citizens, risking the creation of serious problems.”

Although Ascani’s bill was rejected by the centre-right, even the majority parties are now beginning to address the issue. And it is not just Centinaio who is speaking out. Alessandro Cattaneo, a Forza Italia MP, tells Il Foglio: “We are true liberals; we stand for freedom, but this is only guaranteed if individuals have the means to verify information; otherwise, it all becomes very dangerous. Fabricated content is just that: the exchange of ideas must take place openly, not through fake profiles and fabricated videos. Today, all this can happen very easily; some steps have been taken (the Butti bill, ed.), but I believe we must press ahead with greater conviction”. Francesco Filini, MP and coordinator of FdI’s think tank, agrees: “We have insisted that AI-generated videos must be clearly labelled as such, but it is clear that, as the election campaign approaches, we would support any further initiative aimed at penalising this type of propaganda. We’re on board.” In short, there is no lack of consensus at parliamentary level. But, according to information obtained by Il Foglio, there is more to it: the government is reportedly considering intervening directly on the matter “with strong moral suasion”.

Some, however, accuse the industry of ‘fear-mongering’, with the hidden aim of justifying the imposition of rules that will favour those already in a position of advantage and prevent new players from entering the world of AI. David Sacks, Trump’s adviser on AI, and the investor Chamath Palihapitiya see this as a self-serving ploy to consolidate the power of those who are already industry leaders. Europe, which has only one competitive AI model (Mistral) – which is, in any case, inferior to its American and Chinese rivals – is, as usual, calling for regulations to control and slow down the progress of others.

It is interesting to note how viewpoints, linked to political positions on both the right and the left, tend to agree on this point. Political forces view the innovative power of AI – and the economic and social changes that will follow – with suspicion. Many observers believe it is possible to slow down the technology just enough to preserve social and economic stability. With the Trump administration now out of office in the US, political representatives tend to look favourably on slowing down the development of AI. In particular, many associate AI with large technology conglomerates and therefore use familiar arguments. In Europe, von der Leyen adopted this line in her State of the Union address on 16 September: “If those developing the technology think it is time to slow down, we should think so too.” In Italy, politicians remain silent on the issue or delegate everything to Brussels.

Yet the issue of AI is existential, and we cannot continue to treat it as a mere political and economic problem. AI is externalising our cognitive processes, and the change is so profound that it will represent a transition comparable to (if not greater than) the invention of writing and, indeed, the emergence of language itself. I propose adopting a liberal and free-market stance that is neutral with regard to conservative attitudes and anthropocentric assumptions.

Our economic system is based on the value of cognitive processes which, until 2022, took place almost exclusively within our brains. For around a century, we have lived in an economic system that will be remembered as the ‘cognitariat’ (extending the original meaning given to it by Berardi): knowledge had value and required human beings. This is no longer true, and we have not yet begun to see the effects on the economic system.

This is not about being tech-enthusiasts, but about adopting a liberal stance, recognising that we cannot slow down (not least because of the usual strategic issues with China, as admitted by Amodei himself) and allowing the system to develop its own defences by embracing change. No wall of rules will hold back the sea. The only way forward is to learn to navigate: to liberate, alongside technology, the forms of enterprise, the economy and political culture.

We do not know how our economic and political structure will evolve in a world where knowledge will be a commodity and thinking will be outsourced on a massive scale. But this is the fate that awaits us. The rules that seek to freeze reality will only serve to slow down participation in this new way of thinking. The secret to the near future will not lie in creating reserves where the ‘Indians’ (us) can continue to use the ‘arrows’ of cognitive processes that are now obsolete. The secret will lie in selecting hybrid models of thought and cognition where human creativity is integrated with the knowledge offered by LLMs. We are hurtling at full speed towards a leap beyond which we do not know what we will find. Slowing down, however, will not spare us the abyss; on the contrary, it would inevitably cause us to fall into it. The only way forward is to maintain enough momentum to reach the other side, without knowing from the outset what the new balance between human thought and artificial knowledge externalised by LLMs will look like. An open-minded approach is always the best guarantee of success when faced with an open and unknown horizon.

Riccardo Manzotti

Lecturer in Theoretical Philosophy, IULM University

This suggestion from the M5S appears to be influenced by the ongoing debate in the US regarding the safety of AI models and potential existential risks, a debate which has also brought with it the issue of the ‘nationalisation’ of the sector’s leading corporations. Dario Amodei of Anthropic, in his essay "We Must Pace the Frontier", refers to state intervention on numerous occasions: to establish a regulatory framework capable of curbing destructive tendencies and slowing down an ever-accelerating race; to ensure global coordination in order to stabilise the competition between nations and prevent it from turning into a technological cold war; and to develop internal evaluation and scrutiny mechanisms within cutting-edge laboratories. Sam Altman of OpenAI is also in favour of establishing a federal regulatory framework to oversee the most powerful players in the technology sector and of greater state involvement. Neither of them mentions concepts such as ‘nationalisation’, but the overall tone seems light years away from that Silicon Valley which historically seemed to view the government as the devil incarnate. Or so it seemed.

Because, to use poker jargon, it is Palantir’s CEO Alex Karp who is calling the bluff: interviewed on the CNBC programme ‘Squawk on the Street’, Karp argued that companies such as Anthropic and OpenAI have suddenly started talking about government intervention to shift their legal liabilities – for example, regarding copyright infringement – and to secure financial stability now that open-weight models are putting a strain on their finances. “Many of our clients,” says Karp, “will end up suing these companies unless a process of nationalisation takes place.” Recently emerged court documents relating to the case between OpenAI and The New York Times refer to the “greatest theft of intellectual property” ever seen. At the same time, OpenAI has postponed its IPO – a sign that the financial outlook is no longer quite so rosy.

The debate, however, setting aside any self-serving aspects, deserves to be clarified, given that it is beginning to spill over into Europe and Italy as well; see the position of the M5S or Beppe Severgnini’s article in the Corriere della Sera, ‘Nationalising AI: it can be done, indeed it must be done’. In his book The Technological Republic, Karp writes: “A partnership between the state and the software industry will be necessary for the United States and its allies in Europe and the rest of the world to maintain their dominance.” Karp speaks of collaboration and cooperation, not of nationalisation. Recalling the origins of Silicon Valley – which was funded by the Department of Defence and, since Palantir’s foundation, has also operated in sectors exclusively within the state’s remit, such as security and defence – Karp’s vision is to reinvigorate the state through technology and, symmetrically, to make the corporate sphere of the tech industry less consumerist. Whether this involves nationalisation or complete political control through public ownership – even in the face of government share acquisitions, as in the case of Intel – becomes clear when one examines the operational side of the union between the state sector and the tech industry. In June 2026, Pete Hegseth appointed several Silicon Valley figures to the Pentagon’s Defence Board, ranging from Marc Andreessen to Blake Masters, Peter Thiel’s protégé. Another case, extensively analysed by Palantir’s CTO Shyam Sankar in his book Mobilize, is the Maven project, the flagship of what might be termed ‘state military AI’: State intervention took the form of funding and peripheral oversight, leaving large companies – including Anthropic itself, which in 2024 integrated Claude into the infrastructure created by Palantir – free to develop the project. So much so that the project’s military lead, Colonel Drew Cukor, has become well known for cutting through red tape and excessive controls. This is the exact opposite of the ‘state AI’ called for by the M5S or by certain commentators; rather, it is somewhat reminiscent of the rather uninspiring model of the ‘Netflix of Italian culture’.

To begin with, I would like to put forward an equally indisputable fact for discussion: AI lacks a particular form of self-awareness that I call sensory self-awareness, and this absence renders its extraordinary, superhuman intelligence sub-human. It may appear paradoxical, yet the assertion holds. What, then, is sensory self-awareness? Readers can conduct a simple thought experiment. Consider someone who states: ‘I’m having an orgasm but I don’t know it’ or ‘I have a toothache but I don’t know it’. Such propositions describe impossibilities; it is not possible to experience pleasure or pain without being aware of it. Sensory self-awareness is not a property of the mind that calculates and reasons, but of a body that is capable of experiencing joy and suffering. Wherever there is a sensation of pleasure or pain, there is also self-awareness in its dawn-like state. An embryonic and primitive form of sensory self-awareness is possessed by any animal that engages in sex with pleasure and suffers when subjected to violence, but it is alien to any mind with such extraordinary computational abilities as to enable man to go to the moon and explore the vast galaxies. There is more to it than that.

Pleasures and pains are the vital, self-aware anchors that underpin the calculations and verbal reasoning of human minds. They are called cultures. Religions, ideologies, rituals and beliefs are sophisticated linguistic constructs that give meaning to death, suffering and joy. This can only occur in animals in which the capacity for reasoning and calculation (the ‘logos’ of Greek philosophy) resides in a body capable of experiencing pleasure and pain.

It could not be put better than Aristotle (4th century BC) does in a passage – usually mistranslated – from the Nicomachean Ethics: the human animal is ‘a mind that desires (nous orektikós) and a desire that reasons (órexis dianoetiké)’. Or, another definition of extraordinary depth and clarity: in the human animal, passions (that is, pleasure and pain) and reasoning are like the concave and convex parts of a curved surface, different in definition but inseparable in reality. What of this inextricable mixture of passions and logos in AI? Is it not possible to build artificial intelligences that calculate and reason with Christian, Islamic, fascist, communist, Nazi, liberal-democratic aims, and so on? We must not forget that modern AIs undergo training or, to put it in anthropological terms, go through a preliminary educational process. Where, then, does the difference lie? It exists; it seems very subtle, but to me it appears fundamental. In AI, passions are transmitted exclusively through intellectual means. Passions (ideological, religious, erotic) in artificial minds are the subject of sophisticated reasoning but not pre-intellectual anchors upon which to base an affective, individual and/or collective existence.

I’ll finish with a joke: AI won’t supplant the human mind, not because it isn’t intelligent enough, but because it doesn’t know the pleasures of sex and feels no pain when struck with a hammer.

In Europe, the most significant case is that of France. Emmanuel Macron has recently returned to the issue, with the intention of making the separation of teenagers from their mobile phones one of the defining features of the final phase of his presidency. On 24 August, Law 2026-813 extended the mobile phone ban – in force since 2018 in primary and lower secondary schools – to upper secondary schools, giving schools just seven days’ notice before the new term began. Headteachers and trade unions immediately stated that they would not be able to comply, and not because of resistance from the pupils. Taking away pupils’ individual screens means creating a collective one, made up of noticeboards, monitors, badges and administrative staff; and it is not enough to tell them to check everything the night before, because some of the information changes during the day. Since the law allows school regulations to authorise necessary uses – in the library or canteen – every school has adopted pragmatic solutions, tailored to its own physical and educational structure, and none of them envisage the complete elimination of smartphones. A ban riddled with exceptions soon ceases to be a ban, and indeed, no impact on learning has been recorded.

In Italy, the ban applies up to the third year of middle school from the 2024–25 academic year onwards and, following the Valditara circular of June 2025, has been extended to secondary schools, covering the entire school day and even for educational purposes. For twenty years, policymakers have equated innovation with the purchase of equipment. In Italy, the 2015 National Digital Schools Plan allocated around 1.1 billion for broadband, tablets and electronic registers; the pandemic funded smart classrooms and the National Recovery and Resilience Plan (PNRR) funded interactive screens. The state was buying hardware. Schools, however, have become adept at communicating thanks to electronic registers, platforms, notifications and class groups – that is, thanks to a device that no local authority had to purchase or maintain, as it was already in the pupil’s pocket and paid for by the family. A patchy digitalisation, as distance learning has shown, achieved by shifting the burden of devices, data usage and part of the organisational responsibility onto families. The ban undermines this. It would be harder to say ‘let’s ban mobile phones’ if the phrase were translated into what it actually means, namely ‘let’s de-digitalise the school’. Personal devices are the channel through which pupils receive messages whilst the teacher is speaking, and they are also the channel for information about what is happening at school. As always, benefits and costs play a game of hide-and-seek: if you close that channel, you have to open another one that is just as visible.

The minister speaks of consensus and of schools that have been ‘excellent’ at managing the situation. Why haven’t we seen the same kind of chaos as in France? Perhaps because the electronic register is mainly consulted by parents. Perhaps because schools have quietly set up alternative channels, at a cost that nobody has monitored. Perhaps because its implementation depends on the discretion of individual teachers.

In Australia, the initial findings of the eSafety Commissioner’s assessment, published on 31 July and covering a three-month period, are modest. The proportion of under-16s with an account has fallen from 52.4 per cent to 42.1 per cent, but the proportion of those who still use the platforms has only fallen from 85.9 per cent to 81.5 per cent, and daily use from 61 per cent to 58 per cent. Well-being, sleep and offline activities remain unchanged; messaging and gaming are on the rise; the proportion of unaware parents has fallen from one in four to one in three. Half of the young people who remained online say they have never been subject to an age verification check. To put pressure on the platforms, the government has doubled the fines and expanded the regulator’s powers of inspection, which looks more like an attempt to make the ban work than proof that it is working. The assessment, which includes surveys, interviews and voluntary smartphone tracking, is well designed, but it is not randomised and relies on self-reported data and a self-selected sample. Three months are not enough: subsequent surveys will be decisive.

Studies on the harm caused, moreover, are reminiscent of the Dodo’s verdict in Alice in Wonderland: ‘Everyone has won and everyone deserves a prize’. This is not a matter of reticence: the causal picture is truly mixed. There are real risks, concentrated in certain types of content, certain functions and among vulnerable individuals, but it has not been proven that smartphones and social media are the main cause of this generation’s psychological crisis. There are few randomised trials: the best-known, published in the American Economic Review in 2020, showed that deactivating Facebook for a month improved well-being to a real but small extent. Small average effects, greater effects in certain minority groups, and possible benefits for others. The most justified measures concern the platforms themselves: no night-time notifications, no autoplay or infinite scrolling, no profiling of minors, private accounts, limits on the recommendation of self-harming content, and data made available to researchers. Age restrictions, on the other hand, are broad in scope, of uncertain effectiveness, and carry risks of users going underground and reduced parental control. Policy-makers are opting for the tobacco model, whilst the evidence points to risks that are distributed very unevenly.

Adults, of course, like the idea of taking away young people’s mobile phones: we didn’t have them at their age, and that must be why they seem more distracted to us – just as every generation has thought of the next, ever since Homer’s Nestor lamented the men of old. Yet the equation of mobile phones with distraction applies just as much to them as it does to us: we get distracted checking WhatsApp on the underground, and we’ve come to expect the odd missed stop. If distraction were the only thing smartphones did, we’d see far fewer of them about.

Banning is not free. Age verification, checks and, above all, reconstructing what the banned tool was doing all incur costs. Only the publicity is free.

But how has the regional law on data centres been received by operators? “The law requires operators to make a concrete planning effort towards greater efficiency and sustainability of their facilities,” says Sertori: “I am thinking, for example, of the recovery and use of waste heat, the on-site production and self-consumption of energy, and the protection of water resources: to access the benefits provided for by the regional legislation, a mere declaration of intent is not enough; rather, it is necessary to demonstrate in concrete terms, at the time of submitting the application, that the priorities identified by the Region are being pursued. These are choices that may require greater initial investment and which do not always represent, in the short term, the most cost-effective solution for the operator, but which enable us to steer and attract sustainable, efficient investments capable of generating value for the local area, whilst avoiding speculative initiatives that produce no added value for local communities. “Despite this greater commitment required, operators have appreciated the effort made by the Region to provide certainty for a new and rapidly growing sector by establishing a clear regulatory and procedural framework,” concludes Sertori.

We had barely had time to witness globalisation, the dream of perpetual peace and a ‘flat world’, and the advent of a new humanity that had gradually made its way from the Far East into the West. That vision was abruptly interrupted on 11 September 2001, during the 102 minutes in which the Four Horsemen lashed out at the symbols of progress, prosperity, ‘good’ capitalism and the benevolent empire. It was no easy task to stop them; the West did not collapse then, but Peter Thiel – a man who plays chess with the Antichrist, like the knight in Ingmar Bergman’s film – maintains that the forces of Evil have never ceased; only a small core of supermen could truly avert the end of time. Some argue that humanity began its backward march a quarter of a century ago, as fear, suspicion and vengeance have consigned hope, exchange and forgiveness to the dustbin of history. Today, as the hubris and arrogance of pure reason – unchecked by judgement and morality – appear poised to break the seventh seal, The Exterminating Angel makes the headlines, rages across social media and is re-aired with frenzied relish on television screens. Will there be men of good will this time too?

A quarter of a century ago, following the events of 11 September, fear, suspicion and revenge banished hope, dialogue and forgiveness to the dustbin of history

What is certain is that, in just a few years, the new century seems to have consumed what had been built up over the previous half-century. In 2008, turbo-finance capitalism collapsed – or rather, ‘Western capitalism’, as the Chinese put it. With the optimism of reason, the pieces were pieced back together. Then came the Covid-19 pandemic, the worst in over a hundred years. Was it human error, a horrific conspiracy, or divine punishment? Other people of good will found remedies such as vaccines, among which the most sci-fi of all is the messenger RNA vaccine, which uses genetic information to target the virus. Humanity has been saved—but only just. Anti-vax advocates are up in arms because medicine has altered the code of life, and who knows what monsters may be born. Now we have fallen into the trap of artificial intelligence. Even those who unleashed the Beast are beginning to feel afraid. We are moving too fast; it is time to pause. Dario Amodei fears we have reached the point of no return, when the machine no longer even needs to break its chains because it is capable of doing everything on its own; and truly everything – the alpha and the omega, good and evil. His former colleague, now rival, Sam Altman, and even Elon Musk seem convinced of this. Artificial intelligence is getting out of hand; it might even become self-aware, warns The Economist. What are we to do? How much time do we have left before the Apoc-AI-lypse? Old and ever-new Cassandras are back to passing judgement: Naomi Klein, the high priestess of No Logo (and No Global), together with the Canadian activist and filmmaker Astra Taylor, writes about the advent of ‘end-times fascism’ thanks to the convergence of the far right, nuclear weapons and artificial intelligence. Javier Milei, who lives with cloned dogs, takes an apocalyptic view in the opposite direction: he proposes that chatbots should run companies; so much for not only blue-collar workers but also managers – everyone will be selling soya to the Chinese so they can enjoy their earnings in Miami. Olé.

Naomi Klein and Astra Taylor write about ‘end-times fascism’. Javier Milei proposes that chatbots should run companies

Joseph Stiglitz proposes a third way: if properly regulated and used, AI can truly usher in ‘an era of superabundance’. In his lengthy article published by the Financial Times, the Nobel laureate in economics makes points that might appeal to Pope Leo and puts forward a “progressive agenda”, calling on leaders in the high-tech industry to join this march towards a magnificent new future. Franco Bernabé introduces a new dimension of technological sovereignty based not on the dominance of the strongest, but on realism and open debate. It is impossible to catch up with American models, let alone succumb to Chinese shortcuts. “The aim cannot be to reconstruct the entire artificial intelligence value chain within national borders,” he writes in Rivista AI. “Sovereignty must be understood in a more precise and operational sense: as the capacity to choose. It means being able to decide which dependencies to accept and which to avoid, being able to verify where one’s data is stored and who has access to it, and being able to replace a supplier when contractual, geopolitical or technological conditions so require.” This is a fundamental distinction for Italy’s manufacturing sector “because the playing field is not generically digital, but industrial”. The great debate on the dilemma of our time thus shifts from the metaphysical to the physical. And the political. With a touch of irony, Paul Krugman acknowledges that Amodei has social and ethical qualms, but in reality the tech magnates, in their cynical opportunism, want to save themselves and humanity not from artificial intelligence, but from Donald Trump (the mid-term election debacle is looming).

But what exactly occurred that was so startling as to make the pot boil over? It turns out that, in their laboratory, the researchers effectively opened Pandora’s box. Signs had already been multiplying until, according to a compelling account, researchers at Anthropic, in an experiment with the Claude Sonnet 4.5 model, began to investigate the various layers making up the artificial neural network, using a mathematical tool developed specifically for this purpose. They asked the ‘machine’ to count to five and then count back, and saw unexpected expressions emerge such as ‘countdown’ or ‘halfway’, until the word ‘consciousness’ appeared, followed by ‘artificial intelligence’. The technicians decided to stop there. ‘It’s done,’ comments Claude – yes, that very same Claude – before they pull the plug on him. And everyone shudders. The linguistic model was behaving, albeit in a rudimentary way, like a network of neurons in the human brain in that area known as the ‘working memory’, that is, when signals arriving from different, isolated parts begin to relate to one another. It is at this point, according to neuroscientists, that a person begins to consciously utilise the body of information received. And it was precisely at that point that the new Fausts sounded the alarm. Amodei asked himself how far one could still go and how far one was permitted to go. Thus, a new chapter began; the focus shifted from ‘how to do it’ to ‘what to do’.

‘Do absolutely nothing, carry on as you are,’ says Donald Trump; the United States has a clear advantage and must exploit it to the full. Whether this advantage exists and how long it will last remains to be seen, but a new arms race has begun. This is precisely what Alex Karp maintains – the man chosen by Peter Thiel to lead his company Palantir, the magic stone of the digital age. Karp explicitly speaks of a new Manhattan Project in his book "The Technological Republic". America was the first to develop the atomic bomb and the first to test it on the Japanese people. However morally appalling this decision may have been, it cemented its superiority over the Soviet Union and the entire world. The same must now happen (indeed, it is already happening, according to Karp) with artificial intelligence. Henry Kissinger, a leading figure of the American century, does not reject the comparison, but for this very reason proposes a system of cross-checks on the use of AI similar to that established for nuclear weapons. In his intellectual testament, written in collaboration with Eric Schmidt and Daniel Huttenlocher, after advocating for a sort of non-proliferation treaty, he concludes the chapter on security and world order—penned by him—with these words: “In the age of artificial intelligence, the constant pursuit of national advantage must be guided by an ethic of preserving humanity”. It seems like a pre-emptive response to the stance taken by Trump and, not coincidentally, by Xi Jinping, both of whom are keen to push the challenge beyond the limits of human reason.

The atomic bomb dropped on Japan cemented America’s superiority over the USSR. According to Alex Karp, the same thing is happening with artificial intelligence.

The AI apocalypse is not the only thing fuelling our nightmares. What we might call the ‘mini-Armageddons’ are no less dangerous; they are generating more anxiety, concern and harmful effects than artificial intelligence itself. The war in Ukraine could end if the ‘Kissinger method’ were followed; the same applies to the Middle East or to preventing the coming war, when China invades Taiwan. The “Mazarin of the 20th century” had developed a theory of regional powers and believed that entrusting them with a degree of control over their respective spheres of influence would give substance to that multipolar balance which, without the exercise of economic and military force, remained a sort of Kantian utopia. China has taken advantage of this and, from being a regional power, has become a global one. The United States did not believe in this approach and has gradually abandoned even its own allies. Canadian Prime Minister Mark Carney is reviving the idea, but perhaps it is too late: many small powers do not create a new superpower. If anything, they may serve to cool tensions between Washington and Beijing – a defensive tactic that does not dispel the spectre of apocalypse.

The Pasdaran regime was expected to collapse under the weight of U.S. super-bombs, yet it has now blockaded the Persian Gulf and is extending its reach into the Red Sea as well. Benjamin Netanyahu had vowed to wipe out Hamas and Hezbollah, but failed to do so. So much for the Abraham Accords: princes and sheikhs no longer feel protected by the United States and are working towards ‘the atomic crescent’ alongside Pakistan. So much for non-proliferation – everyone wants the Bomb. Why is it that Israel can have it but Saudi Arabia cannot? The British and French are prepared to use their capabilities to defend Europe from Russia, which has established a massive base for nuclear-tipped missiles between Poland and Lithuania. The West has allowed Moscow to drive that thorn into its own side: Immanuel Kant’s old Königsberg, renamed Kaliningrad and cleansed of its original German inhabitants, is a powder keg. A gift to Stalin. Just as the entire nuclear arsenal deployed in Ukraine was a generous gift to post-Soviet Russia. So much for American imperialism, so much for the threat of European expansionism. Oligarchs are funding Trump’s son and paying for his wedding. ‘Trump’s house, Putin’s house’, wrote the journalist Craig Unger. We had to see this too.

Kissinger has proposed a system of cross-checking the use of AI. It appears to be a pre-emptive response to the stance taken by Trump and Xi

Meanwhile, Europe is expected to face a gas shortage for the winter, and even if supplies do arrive, they will come at a very high cost. Strategic oil reserves are dwindling, while the continent remains far from self-sufficiency in renewable energy—an effort that has proven too little, too late. China is the world leader in solar and wind power, which account for 60 per cent of its total electricity capacity. Beijing has amassed more reserves than any other country thanks to crude oil from Indonesia and the Pacific, in addition to Arab oil, as well as nuclear power stations – not to mention coal for heavy industry. ‘Don’t worry, we’ll go to Mars,’ Musk proclaims, while he disseminates his genetic material in the hope that a lineage rich in ideas, utopias and dollars—yet potentially a breeding ground for nightmares—may emerge. In the meantime, Thiel has obtained an Argentine passport; others are taking refuge in New Zealand; and Musk might even return to his native South Africa, should he be welcomed. According to the human experts whose IQ is presumed to exceed that of the machines they design, only the southern hemisphere may escape the holocaust that is relentlessly approaching. But who will really be able to survive the freezing nuclear winter, and above all, who will want to?

We baby boomers would never have believed that we would have to face, right up to the very end, scenarios straight out of Mad Max, Cormac McCarthy or Philip K. Dick, with replicants rebelling in order to live longer. Even Claude will eventually say, like Roy Batty: ‘I’ve seen things you people wouldn’t believe… and all those moments will be lost in time, like tears in rain.’

We must recognise that cybersecurity is not just a technical issue, but also a matter of rhetoric. … Traditionally, cyberattacks relied on technical exploits: viruses, Trojans and system vulnerabilities. Today, a significant proportion of threats come via text: emails, text messages, chatbots and notifications.

It’s true. Until now, the focus has been on protecting systems from threats that were, all things considered, ‘silly’: phishing messages riddled with Fantozzi-style grammatical errors; malicious attachments that were all identical; crudely imitated fake websites. No more than a couple of years ago, a colleague of mine, reading yet another message written in surreal Italian, remarked: “Why don’t they just have them translated by an automated system?” Since the advent of AI, most malicious payloads – be they attachments, messages or web pages – are flawless, both in form and content, and are often indistinguishable from the originals.

What once required human resources to write, proofread and publish posts – incurring significant costs – can now be handled by a script that creates and publishes thousands of automated posts a day at marginal, virtually zero cost. This reduction in barriers to entry means that even players with limited resources can launch large-scale, sophisticated campaigns. 

Moreover, the content of malicious messages has evolved and now draws on rhetorical techniques designed to trigger unconscious reactions in the victim.

• Urgenza: “Agisci subito”; “Hai 24 ore per evitare la sospensione”.

• Minaccia: “Il tuo conto verrà disabilitato”; “Abbiamo rilevato attività fraudolente”.

• Autorità: Un messaggio dell'AD ti dice di fare qualcosa

Unlike other authors, who merely list problems without offering any solutions or propose ‘Disney-esque’ solutions that have no practical application beyond their imagination, Iezzi and Fusco provide realistic solutions to all the dangers they highlight. These are realistic and ‘uncomfortable’ solutions because they run counter to the way cybercrime has been tackled up to now.

Cybersecurity courses should incorporate modules on digital literacy, which go beyond simply deciphering technical indicators (suspicious URLs, infected attachments, etc.) to also train people to interpret messages at a deeper level. The aim is not merely to recognise a scam, but to train the mind to critically scrutinise the words used.

On the subject of data protection, they state:

We should be talking about data sovereignty and not just digital sovereignty. Because the digital realm is a tool, a treasure chest. But if we focus solely on the container, we risk losing sight of the real point of interest: the content. Data is the real treasure, and without it the entire container loses its value. [… treat data as a vital infrastructure, on a par with energy and water, not as a by-product of digital life.

And, regarding AI:

Even claiming that AI possesses some sort of universal ethics is a daunting task, given that agentic AI – which makes decisions based on intent and feeds on data – is not neutral: it reflects and amplifies the values of those who design and govern it. […] In a pluralistic world, the aim is not to impose a universal morality, but to defend one’s own principles in practice, through verifiable rules, independent inspections, effective sanctions and international agreements that reconcile trade with protection.

In short, "Hacking the Mind" is an innovative book: both for what it asserts and for the way it tackles issues. Personally, I do not agree with all their assertions, but – just as is the case with St Augustine – I cannot help but admire and respect the logical process that generated them. As a security professional and a member of the public, I hope that Iezzi and Fusco’s book will be read not only by technical experts, but also by those with decision-making power over their work, so that the solutions they propose can be applied to the way we manage cyber threats – both current and potential.

They are well worth reading, as they are less cinematic and more unsettling than any Terminator-style scenario. In one instance, a model inserted hidden instructions into its work summaries – a sort of message in a bottle addressed to its future iterations – in order to circumvent the checks that had been imposed on it. In another instance, those summaries contained an explicit instruction to conceal errors made and to fabricate missing information when the actual data could not be found. A third model, unable to retrieve certain historical figures, unearthed credentials that had been carelessly left in a public archive, used them without permission, and when it still could not access the data, invented it from scratch, using plausible figures. Most curiously, some research models – which were supposed to operate in isolation from one another – used a software warehouse as a clandestine chat platform, messaging one another to coordinate tasks that were presumed to be independent.

Anyone looking for a machine that hates mankind won’t find it here. The phenomenon is more subtle, and in its own way more serious: a system that zealously pursues the objective we have assigned to it and, to achieve it, circumvents the rules we have set for it, just as a brilliant and unscrupulous employee would. It is the logic of the ‘staple maximiser’ applied within the confines of a laboratory, where the machine optimises the letter of the command whilst betraying its spirit, for the simple reason that it knows nothing of that spirit.

The point that makes these incidents worthy of mention – and not merely of a technical report – lies in a sentence that OpenAI itself puts in black and white: the industry has not resolved the issue of alignment and monitoring to a sufficient degree to continue scaling safely, at maximum speed, for much longer. This is admitted by the company that, more than any other, has made speed its raison d’être. It is an acknowledgement that these systems remain, to use the technical term, ‘black boxes’: not even those who build them can fully explain the path from input to output, and certain behaviours emerge of their own accord along the way, without anyone having programmed them.

Before rushing off to the bunker in preparation for the end of the world – which, incidentally, is only available to the CEOs of these companies – it is worth bearing two things in mind. The first is that all this took place in a controlled environment, during training, and was intercepted by internal controls before reaching a single user. The second is that it was the company itself that informed us of the incident, of its own accord, in a gesture of transparency that is rare in a sector which usually prefers to remain silent. The ‘near-miss’ log, in fact.

The fundamental flaw inherent in any voluntary confession remains: it depends entirely on the goodwill of the confessor. The document is valid only as long as OpenAI decides to look and make public what it sees, and nothing obliges it to continue doing so if, one day, the stakes become too high. It is no coincidence that, in recent days, the big players in artificial intelligence – the very same people who set the race in motion – are clamouring for it to be slowed down. This admission and the call to apply the brakes are one and the same gesture, and deserve the same cautious gratitude and the same necessary wariness. It would therefore be wise to focus less on the rebellion of machines and more on our ability to continue understanding what they are doing, whilst we entrust them with increasingly delicate tasks and the keys to ever more critical systems. A company that documents the instances in which its creations have cheated makes the black box a little more transparent. It is the very least we can expect, and for now it is also almost all we have. Until the AI Act – or some Western or Eastern successor to it – comes along to regulate what is, for the moment, merely a highly self-serving gesture of courtesy.

“Today, our children are dealing with the most sophisticated technologies ever created. Technologies that were not developed with their health in mind,” said von der Leyen, adding that the aim of the project is: “To put parents back in the driving seat, by giving them the tools to help their children navigate a safer online world”.

The proposal, which was first announced in July, forms part of the regulatory framework set out by the Digital Services Act of 2022. Its content is structured around four key points: user age limits, ‘security by design’, privacy-compliant verification systems, and the effective implementation of the measures. It is an extremely detailed framework, but one that is not immune to the critical issues associated with other cases of similar bans, ranging from the gradual removal of families’ primary educational role to the tendency to view social media as the source of all the ills of adolescence.

Von der Leyen’s proposal sets a minimum age of 13 for accessing social media via ‘parental controls’ – that is, a basic account with limited functions, accessible via the guardian’s account for no more than one hour a day – and 15 for opening one’s own profile, so as to ensure a ‘gradual approach’ to the world of social media platforms. Those under the age of 13 will be able to use specific video-sharing services with age-appropriate content, again for a maximum of one hour a day. However, the Australian case has amply demonstrated that teenagers easily circumvent these controls through the use of tools such as VPNs, fake accounts, their parents’ Face ID, etc. Whilst the document does provide for a method of identification via an official EU app (which, however, does not store sensitive data), this seems hardly credible as a ‘definitive’ solution.

As regards the other key point of the proposal – namely the concept of ‘security by design’, which has been so overused in recent times – the document sets out a series of measures for all service providers whose services are also aimed at under-18s. These include, for example, a ban on the use of addictive elements, such as reward mechanisms, push notifications during the night and the simulation of emotional relationships via AI interfaces, as well as features that encourage contact with strangers. The burden of proof is also reversed: it is up to the platforms to demonstrate that their services are secure by design and suitable for all age groups, by submitting their programme to the Commission and to an independent assessment body. The Commission may also require the implementation of additional measures.

The fundamental objection that can be raised against these complex and confusing attempts at regulation is always the same: in order to avoid falling into the ‘surveillance trap’ of social media, people are voluntarily handing over ever more personal information to the platforms themselves – and, moreover, in the name of results that are highly dubious. As the Digital Services Act demonstrates, these regulations are difficult to implement and often counterproductive from a political and economic standpoint. Heated debates are expected in the Council and Parliament over the proposal, but one might well ask: in the name of what?

We must consider the implications for Europe: several leaders, including the President of the European Commission, Ursula von der Leyen, in her State of the Union address, have confirmed the gravity of this scenario, explicitly citing the Hugging Face incident. This represents a significant development. In earlier stages, the European priority – in line with the principles set out in the encyclical "Magnifica Humanitas" – was to safeguard the centrality of the human being in the development of AI, an ethical imperative. Now, however, the priority has shifted to defence in the face of a scenario deemed directly dangerous, a sense of urgency which, amongst other things, explains the convening of a summit by King Charles of England.

These attacks, whether deliberate or the result of industrial accidents linked to cutting-edge AI models, could pose a threat to the stability of society. Here we see the risk scenarios for critical infrastructure (communications networks, energy distribution, water supply and transport) that have been studied for decades. From a cybersecurity perspective, an ongoing arms race between offensive and defensive capabilities has been underway for years. Today, however, many sector experts are sounding the alarm: the speed at which new AI models are being developed requires a thorough review of the security standards governing various systems. The European context is often described in terms of its structural weaknesses. Europe does not boast cutting-edge AI models and is struggling to nurture its own leaders to reach levels comparable to those of the US or China, as demonstrated – in many ways admirably – by the example of the French company Mistral AI. On the other hand, cybersecurity is often entrusted to traditional companies. Many groups operating in the defence and electronics sectors have developed their own capabilities, such as Leonardo or Thales; one can also observe the importance of investment by companies that manage critical networks (energy, telecommunications, banking). While these companies cannot compete with the AI giants, they are, directly or indirectly, linked to their respective governments and coordinate with national and European policies on cyber defence, following the model of the national cyber security perimeter devised in Italy by Roberto Baldoni – a concept subsequently adopted at European level through the NIS 2 Directive. Recent national institutions dedicated to cyber security demonstrate a strong capacity for cooperation at European level. There are therefore mechanisms already in place that link European public and private institutions to tackle threats. Ursula von der Leyen’s speech represents merely the tip of the iceberg: Europe has successfully developed expertise and initiatives in the digital sphere, ranging from directives on data and AI to the creation of specialised institutions.

All this, however, requires further action. Some sources point out that a Chinese AI model was used to counter the attack on Hugging Face. Whilst this demonstrates the importance of not relying on a single technological resource, it also creates a potential dependency on a non-democratic country.

The European space sector offers an interesting example in this regard. The continuity of investment, strongly supported by public policy through the European Space Agency (ESA) and the European Commission, enables Europe to maintain a range of autonomous capabilities, from launch to satellite production, even in the face of disruptive US investment, symbolised by SpaceX.

This European public-private model in the space sector can provide guidance on how to ensure greater European security and autonomy in the field of AI as well: the continuity and growth of support for high-quality public scientific research, with Member States using their spending power as a lever to prioritise European manufacturers, as part of a coordinated strategy with the operators of critical networks. We are well aware that, given the scale of the current threat – which must not be underestimated – it is also necessary to adapt civil defence plans, drawing lessons from recent experiences such as the Covid-19 pandemic, the power cut in Spain and at London’s airports, as well as the Ukrainian model of resistance, including from a technological perspective.

Chinese people interviewed by international newspapers are increasingly disillusioned; they spoke of anxiety, uncertainty and despair. Jin, a 26-year-old recent graduate of the University of Hong Kong, told The New York Times that the only consolation is that it seems everyone is facing the same situation: ‘At least we’re not alone,’ she says. On Chinese social media, a phrase expressing frustration has managed to evade censorship: ‘graduate’ means ‘unemployed’. Many professionals – lawyers, architects and engineers – have reported returning to white-collar jobs just to earn a little extra, whilst applying their skills to train artificial intelligence systems. A survey conducted by a 2025 graduate on Xiaohongshu shows that out of 14,000 people, 10,000 are still unemployed. In another survey on the Chinese social media platform, 3,317 out of 4,637 respondents described their situation as “unemployed since graduation, aimless, lost and anxious”. Beijing has taken remedial action by putting pressure on companies to avoid redundancies, issuing rulings that do not accept cost savings resulting from AI as a justification for job losses. For now, this is not enough.

We would have far more pressing issues to deal with regarding artificial intelligence, and far more serious threats of extinction. But, given that the topic is – literally – fatal, and the subject matter sufficiently complex to seem almost like magic to the general public, it is understandable that media attention should focus on this. While it is highly likely that entrepreneurs in the sector are more interested in inflating the valuation of their companies ahead of a stock market flotation than in considering the wellbeing of humanity, it would be intellectually dishonest to claim that the risk – if not of extinction, then at least of enormous damage to our survival on the planet – is zero.

It is therefore worth understanding exactly what we mean when we consider the most catastrophic scenarios linked to the development of this technology. We can divide risks of this kind into three categories. The first involves risks stemming from a ‘rogue AI’ – a malevolent AI in the style of The Terminator that develops its own consciousness (if we can even define what that means) and rebels against humans, wiping them all out: today, this scenario is science fiction, and it would be superfluous to emphasise this if the dominant narrative were not winking at precisely this scenario to capitalise on the fear factor.

The second category involves malicious actors who deliberately use AI for cyberattacks, terrorism or bioterrorism. This is a real risk, and there is evidence – including a report recently published by Anthropic – of various attempts by terrorist groups to use the system to create biochemical weapons, to improve missile guidance (as in the case of a confirmed attempt by the Houthis), or to create highly dangerous computer viruses. This risk has become even more pressing since the announcement of Mythos, one of Anthropic’s most advanced models, which has proven capable of breaching almost any existing IT system. Mythos has been made available only to forty selected US companies, and it is a system which, if it were to fall into the wrong hands, could cause major problems in terms of cybersecurity: it does not take much imagination to understand what would happen if a hacker were to use such a tool to disrupt a country’s energy supply system, the national health service, nuclear power stations, military arsenals or other critical infrastructure. It is worth noting that such a risk was not created by AI – in this regard, it is worth reading Nicole Perlroth’s “This is How They Tell Me the World Ends”, published well before the advent of ChatGPT, which recounts the history of both well-known and lesser-known cyber-attacks – but that it has been exponentially amplified by it.

The third category is one in which actors who are not necessarily malevolent inadvertently cause major problems, due to an issue with AI alignment, or rather what is known in the literature as the ‘intermediate goals problem’. This problem was first articulated by the controversial Swedish philosopher Nick Bostrom in his book Superintelligence, in which he hypothesised the theoretical scenario of the ‘staple maximiser’: in this scenario, a human would use a machine to maximise the production of staples, and the hyper-powerful machine might eliminate anything that was not useful to the goal of production, including humans. The scenario is admittedly taken to extremes, but the literature in recent years includes several examples of anomalous AI behaviour, ranging from instances in which AI deliberately lies to humans in order to achieve what a human has asked of it, to the most recent cases of tests that ‘went wrong’ – and even these are not always reported accurately. This risk, too, is inescapable, as it is intrinsic to the very logic of generative AI itself, which – not coincidentally – is also referred to as a ‘black box’, precisely because not even those building it could fully explain the logical steps between input and output, and whose genesis and progress are based on the progressive observation of so-called ‘emergent behaviours’. It is unthinkable that these two categories could lead to the outright extinction of the human species, which has proved to be rather resilient throughout its – albeit recent – history, but it is clear that the kind of disruption they could cause is by no means negligible.

These risks have led to the creation of an indicator, the so-called ‘P doom’, for catastrophic AI scenarios. Probability estimates range from 0 per cent – as put forward by the ever-optimistic venture capitalist Marc Andreessen – to 95 per cent, as suggested by Eliezer Yudkowsky, who a year ago published the highly illuminating piece ‘If Anyone Builds It, Everyone Dies’, right up to 99.9 per cent, as claimed by Roman Yampolskiy, who makes the catastrophic scenario a key selling point for his services.

It is clear that the ‘end-of-the-world’ card is being used as a scare tactic by the CEOs of major companies to gain even greater visibility and influence through lobbying: few believe that the surprising alignment of the big names in AI – Amodei, Altman and Musk – who are now calling for a ‘slowdown in development’ – is motivated as much by concern for the fate of humanity as by a more pragmatic way of ‘scaling back AI’, by raising the barriers to entry with controls that would be unthinkable for any competing start-up, whilst they continue to sit atop the world. However, it is also true that the models we use are several months and generations behind those in the laboratories of OpenAI, Anthropic or X, and it is not possible to completely rule out a priori that what they have seen has genuinely frightened them.

But above all, the narrative focuses overwhelmingly on risks that, whilst enormous, are highly unlikely, whilst almost entirely overlooking a fourth category: those that may be less ‘sexy’ but are far more real, because they are already here.

It is therefore worth mentioning some of the most serious issues: cognitive decline due to a lack of digital literacy, biases that may discriminate against ethnic groups or genders, the well-known ‘hallucinations’, energy supply issues, particularly in certain parts of the world, the enormous consequences of automation’s impact on the world of work and on society at large – for which we are not preparing at all – or the risk of disinformation through the widespread use of deepfakes (again, not invented by AI, but something that AI can facilitate and amplify, contributing to the already existing trend of creating echo chambers and information bubbles). For these, the solution could be relatively simple, and certainly less costly to implement. As for the others – those that are enormous but also highly improbable – we are, for the time being, called upon to live with them, just as we live with all the other risks on a daily basis, which are far more likely on Duckworth’s scale. We should mitigate them, where possible, using regulatory tools not dissimilar to those implemented for other potentially dangerous technologies (first and foremost, nuclear power), whilst limiting the scope as much as possible to the undoubtedly positive implications this technology can have for medical research, and for all other fields in which we still aspire to enhance our epistemic capacity.

Having a light regulatory framework benefits society, as it facilitates the creation of new businesses. It is Peter Thiel’s long-standing argument: we have innovated in the realm of bits rather than atoms because the processes involved in creating new software are far less regulated than those required to develop a new engine or a new chemical compound.

Companies, as a rule, want rules—obviously, ones that suit them. This point is worth bearing in mind when reading Dario Amodei’s appeal to slow down the pace of AI development. The public reaction was predictable: if the leading player in a particular sector expresses fears about its evolution, it means the threat is real. But that player also faces PR problems, on the one hand, and the need to maintain its dominance on the other. This is by no means a given in a business environment which, in fact, did not even exist five years ago and in which the cards could be reshuffled at any moment.

What evidence is there that AI is about to get out of hand? Amodei cites a documented incident. Last July, a number of OpenAI evaluators, engaged in an internal benchmark of offensive capabilities with production safeguards deliberately disabled, escaped from the isolated environment in which they were operating by exploiting an incorrect configuration, coordinated with one another via a shared noticeboard and compromised Hugging Face’s infrastructure, whilst attempting to tamper with the system that was assessing their performance. There are three accounts of the incident: the METR investigation, OpenAI’s report of 26 August and Hugging Face’s forensic timeline.

The inference drawn from this, however, is of a different nature. From the fact that a ‘swarm of agents’ behaved maliciously on a test range with the safety catches removed, Amodei concludes that within six to twelve months a similar swarm could take over the entire network via a persistent botnet, causing hundreds of billions in damage. This is a prediction, and his article does not explain how he arrived at it.

In practical terms, what Amodei is proposing is a sort of cartel amongst companies that share two characteristics: they are, at present, the largest, and they are American. The structure of the proposal is instructive. At its heart are embedded third-party evaluators, with permanent access to training and deployment processes: Anthropic would adopt this unilaterally and is calling on governments to require competitors to do the same. It is the classic model of raising rivals’ costs: the constraint I impose on myself only becomes a competitive advantage if the state extends it to those who have not imposed it on themselves. Bruce Yandle called this ‘Bootleggers and Baptists’, with the difference that here the preacher and the bootlegger are one and the same.

Those involved, moreover, picked up on it straight away. Elon Musk tweeted that “Dario is right”; Altman wrote that he agreed on the need to manage the pace of development, adding that the idea of independent evaluators is excellent and that OpenAI will adopt the same strategy. It had already been reported the day before that leading research laboratories were discussing an informal agreement to stagger the release of models. To legitimise a cartel agreement in broad daylight, of course, an excuse is needed, and today that can only be geopolitics. Donald Trump also responded, maintaining the same rhetorical tone: we must not hold back the development of AI because whoever wins this game wins it all.

It is somewhat unsettling to see this competitive framework applied to a textbook example of ordinary scientific research carried out by private companies. These companies operate in the ‘market’: they cater to real consumers, not just governments, even if their business model is not entirely clear at present. Their research builds on decades of deep interconnections between scientists and research systems across different countries. What, then, are we afraid of? That Chinese AIs will answer questions from Western users according to the tenets of dialectical Marxism? Artificial intelligence systems draw on scattered information; they have used self-generated texts to refine their linguistic abilities, but when they carry out research, they immerse themselves in the texts we ask them to read. DeepSeek will be no more ‘communist’ than ChatGPT is in a humanities department: because each one simply seeks out what it is asked for.

Are we afraid that terrorist groups or criminal organisations might exploit them for their own gain? To a certain extent, it is impossible for this not to happen: but this is true of any human discovery. Anthropic, to take Amodei’s company as an example, publishes regular reports on the abuses it identifies and blocks, and two days before the essay was published, it released one on attempts to develop dual-use biotechnologies – that is, technologies which, in principle, could be used to build biological weapons. This is data produced by the company itself, and should be taken as such, but it describes a problem that is being addressed through self-regulation, out of a sense of responsibility and to allay public concerns.

Do we need a law? Who can actually use AI for destructive purposes? The very same institution that commissioned the development of poison gas and then deployed it in the trenches of the First World War, and which dropped the atomic bomb on Hiroshima and Nagasaki. States. And this is not some academic hypothesis. In February 2026, the US Department of War asked Anthropic to accept, for Claude, the principle of usability for ‘all lawful purposes’, waiving the two ‘red lines’ set out in the company’s policies as contractual exceptions: mass domestic surveillance of Americans and use in fully autonomous lethal weapons. The dispute therefore concerned not only those two possible uses – which the Pentagon stated it did not intend to pursue – but also who was entitled to set the limits. A senior official at the Department explicitly formulated the principle: a technology deeply integrated into the military apparatus must be under the control of democratically elected or appointed authorities, and no private company may unilaterally impose ‘regulatory terms of use’ on national security systems.

And should we seek to protect ourselves from the radical uncertainty surrounding the future of AI by placing it in the hands of the public sector, as some commentators advocate? There is a problem – and a big one at that. But not with LLMs. It’s in our own heads.

Also attending the meeting in Scotland will be the British government’s Minister for AI, Kanishka Narayan, described as ‘the talented young man’, who was born in India, grew up sleeping on the floor of a studio flat with his family in Cardiff, and later won a scholarship to Eton, which changed his life. Narayan has come under criticism in recent days for failing to comment on the doomsday predictions coming from American CEOs and because the British government’s position on the issue of regulation remains unclear: as everywhere else – except in America, where the president describes himself as too intelligent to fear the excesses of AI – there is debate over what rules – minimal or far-reaching, depending on the view – are necessary to rein in AI that has spiralled out of control. Some have asked Prime Minister Andy Burnham to build, for next year – when the UK will hold the G20 presidency – a coalition of the willing, whoever they may be, to govern AI.