The real risk with AI is not taking any risks at all

The ‘Oppenheimer moment’ in artificial intelligence – that precise moment when those creating something innovative ask themselves whether the benefits of that technology outweigh the harms – is an interesting test for observing two categories of commentators who generally feed off one another, and not just when it comes to AI. On the one hand, the theorists of the inevitable apocalypse. On the other, the proponents of inescapable optimism. The story is one you’re familiar with. Some of the leading figures in the AI industry (Amodei, Musk, Altman) have sounded a high-profile alarm: artificial intelligence can have disastrous consequences, and for this reason we need to pause for reflection before bringing about the end of humanity in a rather hasty manner. What is notable is that the very champions of techno-optimism have become prophets of doom. This contradiction has emboldened the catastrophists to lump everything together. The weekend’s final verdict: AI is a danger, let’s slow down, and if possible, let’s get rid of it before it’s too late. But is that really the case? The issue at hand is not whether artificial intelligence needs limits. The real issue is whether those shrewd operators who run the American AI sector have seized the opportunity to try and align themselves – not least for marketing reasons – with the ‘good guys’ (which can be beneficial in times of anti-AI sentiment), whilst at the same time attempting to ask the US government for an exemption from competition laws so they can be authorised to set common safety standards and limits on the pace of progress.

Jim Cramer, a well-known presenter on CNBC, and Aidan Gomez, CEO of a major AI company, have attempted to explain in simple terms another key point: why the Big Tech giants have a little-discussed interest (which they disguise as an ethical appeal) in promoting a slowdown in AI. The problem for Big Tech firms involved in AI is that they have been spending heavily on AI for years without managing to generate the hoped-for profits. And having an (ethical) excuse to tackle an industry-wide problem can prevent them from finding themselves in a situation where one company falls behind whilst others forge ahead. Key figures. In 2025, OpenAI (Altman) generated revenue of around 13 billion, but incurred total costs of around 34 billion, of which around 19 billion was spent on research and development. In the first quarter of 2026, X.AI (Musk) had revenue of around 818 million, but spent 2.4 billion on research and development alone. In 2026, Anthropic (Amodei) finally began to post an operating profit: around 559 million in the second quarter. But with an imminent IPO on the horizon (expected by the end of 2026), the prospect of not being forced to immediately reinvest every dollar earned in the extremely costly race to develop the next model could be a godsend. Setting limits on technology, as the President of the Republic, Sergio Mattarella, pointed out yesterday in his birthday message to Pope Leo XIV, is important. But to do so requires more than just feelings, intuitions and vague figures (regarding the estimated 10 per cent probability that AI could cause the extinction of humanity, we would need more than just a tweet). Transparency is needed, as they say, in highlighting errors and abuses, and sharing them when they arise (which does not happen today). But we must also remember that jumping on the ‘stop AI’ bandwagon does not mean reducing its risks. It means slowing down research into new medicines, depriving businesses of opportunities to develop their expertise, and creating competitive advantages for non-democracies that show no sign of stopping. It means failing to understand that today there is only one risk greater than taking AI-related problems lightly: ceasing to take risks by giving up on building artificial intelligence capable – with the right safeguards, without ideological doom-mongering or irresponsible optimism – of improving our world.