Ex-OpenAI researchers and AI firms give MPs evidence on AI risk and regulation

Right · 1Daniel Kokotajlo, a former OpenAI researcher, told the Business, Innovation, Science and Trade (BIST) Committee on Tuesday that "the risk tolerance across the industry is catastrophically high". 1 Right · 1Kokotajlo said he quit OpenAI after concluding it "would not behave responsibly around the creation of this technology". He also said he felt pressure to "mince my words a bit" in public communications and that some internal research and forecasts were stopped from being published, according to City A.M. 1 Left · 1Wired reported that Kokotajlo also said "I don't think the companies can be trusted to self-regulate". 2

Mixed · 2Steven Adler, who left OpenAI in late 2024, said there had been resistance inside the company to calling its technology "dangerous", though he acknowledged its culture may have changed since. He told MPs he was "not confident in the safety culture across the industry" and urged governments to "set the ceiling" on the risk companies can impose. 1 2 Right · 1Geoffrey Irving, a former researcher at OpenAI, Anthropic and the UK's AI Security Institute (AISI), said capability advances were far outpacing safety work. Asked about his earlier estimate of a 50 per cent chance of advanced AI destroying humanity, he said a "coin flip is roughly the right model to have"; City A.M. noted this is a personal assessment rather than an established scientific probability. 1 Left · 1Irving said the labs believe they are locked in a race that governments will not interrupt and that they cannot leave. "Both of those are false," he said. "They could just stop." 2

Left · 1The session also heard from four frontier AI firms, four former staff and senior civil servants. Anthropic's head of Europe, Liam Booth-Smith, said "the rules should be binding" and that voluntary commitments are not enough. 2 Left · 1OpenAI vice president Tom Duff Gordon backed a statutory framework and an international convention, saying "there should be some elements that should be mandatory", starting with incident reporting. 2 Left · 1Google DeepMind's Owen Larter said the focus should be on standards, and Meta's Rob Sherman pointed to existing regulators, telling MPs they were "not writing on a blank page". 2

Left · 1Civil servants told the committee the current system is voluntary. AISI director Henry de Zoete said the institute relies on company and foreign-government cooperation for access and cannot stop a release, and Cabinet Office top civil servant Cat Little answered "At the moment, yes, it is" when asked if the regime was voluntary. 2 Mixed · 2British access to models runs through Washington: OpenAI's Duff Gordon said it provides models pre-deployment to the US counterpart CAISI and will continue with AISI as long as the US government allows, and Booth-Smith said Anthropic could not initially give AISI full unrestricted access to Mythos 5.1, which he called "a question for the US government". He added that AISI now has access, and City A.M. reported Washington restored UK access after a temporary restriction. 2 1 Left · 1Committee chair Liam Byrne told Wired the aim was to show "that you couldn't rely on voluntary self-regulation" and to expose differences between some of the wealthiest firms. He said the committee would examine the UK's path as a middle power and wants ministers to use Britain's coming G20 presidency to seek an international convention not dependent on the current White House. 2

Left · 1Nearby, MI5 director Ken McCallum used his annual threat update to name AI's potential to change the stakes on bioweapons as the misuse case to which he attaches greatest weight, though "not yet realized". 2 Right · 1City A.M. added context that AI minister Kanishka Narayan warned on Monday that a major breach of Britain's cyber defences was likely inevitable, and that Anthropic warned its models may soon self-replicate and that its IPO filings reportedly cite "existential risks to humanity". It said OpenAI and Anthropic had been contacted for comment. 1