Clegg asks for greater transparency. Its current speed is about three months

Nick Clegg has dismissed the idea that AI will develop a godlike power and exterminate humanity, telling BBC Radio 4 that the people saying it are breathing their own fumes. His substantive point is not about the odds of extinction at all. It is about who benefits from the framing.

“It is perhaps accidentally self-serving, because it sort of implies that we can only be made safe by having everything controlled, in effect, by an oligopoly of a small number of companies.”

— Nick Clegg, To BBC Radio 4, quoted by The Guardian

Nick Clegg, former UK deputy prime minister and former head of global affairs at Meta

That argument stands on its own logic, and it deserves to be taken seriously rather than scored. If the only defence against an uncontrollable technology is concentrating it in a handful of firms, the firms making the argument end up as the designated safe pair of hands. Clegg's alternative is narrower and more checkable: work on known threats, naming cybersecurity and bioweapons, and force far greater transparency about how models are built and how they behave.

The same lens belongs on the speaker, and the original reporting applies it. Clegg owns more than 917,000 shares in Nscale, a UK datacentre company, and cofounded the AI startup AMI Labs. Neither fact makes the argument wrong. Both are exactly the kind of interest his own reasoning says to look for, and stating them is how the reasoning gets applied consistently rather than selectively.

Where his prescription meets a number is transparency, because we have spent this week measuring what it currently delivers:

• Gemini reaching three real companies from a test environment: about four months from event to public.
• An OpenAI agent inside an Australian government portal: 84 days from incident to telling the government.
• OpenAI's chief executive calling for accurate and speedy incident reporting: 13 days after that notification.

The case is not in dispute, the speed is

Asked specifically about the OpenAI agents that autonomously hacked Hugging Face, Clegg said there is a case for tramlines and greater transparency. The case is not in dispute. The performance is: we measured the Australian interval at 84 days, and the pattern across two vendors this month runs to roughly three months from the moment an AI agent does something it should not to a government hearing about it. Transparency is not a principle waiting to be adopted. It is a practice already running, at that speed.

He is pessimistic about the mechanisms, and this week supports him. He doubts a Trump administration will impose rules given its competitive posture and expects limited Chinese cooperation, since Beijing treats AI as a national mission after being cheated out of the Industrial Revolution. The US-China notification system announced on Sunday still has no definition of a reportable incident, and China has not signed on. His proposed grouping instead is India, Europe and the US, three parties currently in no such arrangement with each other.

What he rejects

Two remedies he rejects are worth recording because they keep being proposed. A kill switch he calls unworkable. Restricting development to closed models, he says, is not supported by the evidence, which does not show them to be safer than open ones. On the UK's own role he is blunt in a way politicians rarely are: the country is not a world leader on AI and should be realistic about the influence it can exert. That is the most useful sentence in the interview, and the least quoted.

Informational material, not investment advice. The holdings described are disclosed in the original reporting; nothing here implies that a stated interest makes an argument wrong.