Government admits: the entire UK AI oversight regime is voluntary
Two of the world's leading AI labs asked a British parliamentary committee to regulate them by law – while Google DeepMind and Meta declined. The same day, officials admitted the entire UK oversight regime is voluntary, and MI5 warned of AI-enabled biological weapons.
On Tuesday, October 13, 2026, the Business, Innovation, Science, and Trade Committee, a cross-party committee of British Members of Parliament, held an evidence hearing featuring four frontier labs, four former employees acting as whistleblowers, and senior British civil servants. The result was a rare public split among the largest AI companies over whether voluntary commitments are sufficient – and an unexpected fragility in Britain's AI safety regime, laid on the parliamentary record.
Two companies say: bind us
Anthropic's Europe head Liam Booth-Smith was the most unequivocal. Asked whether the company believes AI safety now requires a statutory framework, he said, according to WIRED: "The rules should be binding." The company has, he said, "consistently believed that there should be binding rules for the companies, and that voluntary commitments are not enough."
OpenAI was more cautious, but also backed legislation. Vice President Tom Duff Gordon advocated a statutory framework and an international convention, saying that "there should be some elements that are mandatory" – with mandatory incident reporting as a starting point. "We don't think that basing everything on voluntary commitments for frontier risk is a good long-term plan," he said, according to WIRED.
The other two say: not yet
Asked directly about support for a statutory framework, Google DeepMind's Owen Larter responded: "Focusing on standards is what we would do." Meta's Rob Sherman pointed to Britain's existing regulators and told members they were "not writing on a blank page."
This was precisely the split that committee chair Liam Byrne said he wanted to expose. The committee aimed to "shine a light on the big differences in opinion between some of the richest companies on earth," he told WIRED after the hearing, and to show "that you cannot rely on voluntary self-regulation."
Former employees: they cannot be trusted
The hearing also heard from four former researchers at the labs, including OpenAI's former employee Daniel Kokotajlo. "I don't think the companies can be trusted to self-regulate," he told members. However, in the available coverage, the testimony of Steven Adler, another of the whistleblowers, is only partially documented, so the full extent of this portion of the hearing is unknown.
Officials admit: it is all voluntary
Perhaps the heaviest moment came from the government's own side. Cat Little, the most senior civil servant in the Cabinet Office, which houses the AI Safety Institute (AISI), was pressed on whether the entire regime was voluntary. "At the moment, yes, it is," she replied, according to WIRED.
AISI director Henry de Zoete admitted that the institute is entirely dependent on the cooperation of companies and foreign governments to gain access to models – and that it cannot block the rollout of a model it considers dangerous.
The dependency extends across the Atlantic. OpenAI gives its American counterpart CAISI pre-deployment access to models, and Duff Gordon explained the arrangement: "We give our models pre-deployment access to CAISI … We will continue to give our models pre-deployment access to AISI in the UK … and as long as we're able to do that under the US government, we will continue with that." Anthropic stated at the hearing that it initially could not give UK AISI full, unrestricted access to the model Mythos 5.1.
Same day: MI5's bioweapon warning
The hearing was accompanied by a separate but striking event the same day. In its annual threat assessment, MI5 director general Ken McCallum warned that AI's greatest misuse risk is its potential to "change the calculus on existing or even new biological weapons" – "not yet realised," he added. According to POLITICO, MI5 has established an AI Threat Assessment Centre (AITAC), drawing on AISI, the National Crime Agency and the National Cyber Security Centre, to help the government assess potential threats.
The timing is context, not causation: the hearing and MI5's warning were independent events that occurred on the same day, but together they underscore what is at stake in the debate over whether Britain's voluntary approach is adequate.
What remains open
All the quotes above come via WIRED and POLITICO coverage, not a primary transcript of the hearing, and cannot be independently verified against the parliamentary record. It is unclear what recommendations the committee will make, and whether any of them will lead to legislation. Several details – including the full circumstances of Anthropic's restricted access to Mythos 5.1 and the rest of Steven Adler's testimony – remain unclear.
But what is now on the record is remarkable nonetheless: two companies building the technology asked to be bound by law, two refused, and the government admitted it cannot compel anyone to do anything.