OpenAI and Anthropic Ask UK Parliament for Binding AI Laws

Two of the world’s leading AI companies went to the British Parliament this week and asked to be regulated. Two others refused.

The split played out on October 13 at a hearing of the House of Commons Business, Innovation, Science and Trade Committee, chaired by MP Liam Byrne. The committee called in four frontier AI companies — Anthropic, OpenAI, Google DeepMind, and Meta — along with four former lab researchers turned whistleblowers and senior British officials. According to Wired, which reported the hearing on October 14, the point was to expose just how far apart the industry’s richest companies are on self-regulation.

Who wants the law, and who doesn’t

Anthropic was the most direct. Liam Booth-Smith, the company’s head of Europe, was asked whether AI safety needs a binding legal framework. “The rules should be binding,” he said. He added that Anthropic has consistently believed there should be enforceable rules for companies, and that voluntary commitments are not enough.

OpenAI was more cautious but landed in the same place. Tom Duff Gordon, the company’s vice president for Europe, the Middle East and Africa, said he supported legislation and an international convention. “There should be some things that are mandatory,” he said, starting with mandatory incident reporting. Basing everything on voluntary commitments, he said, “is not a good long-term plan” for the risks of the most advanced models.

The other two did not follow. Owen Larter, a senior director at Google DeepMind, was asked directly whether he supported legislation and answered only that the company’s focus was on standards. Rob Sherman, Meta’s global policy vice president, said Britain’s existing regulators were capable of handling AI, and that lawmakers were “not writing on a blank sheet.”

The committee chair told Wired the aim of the session was to make the disagreement visible — and to show that self-regulation cannot be relied upon.

Britain admits it can’t stop a dangerous model

The more striking testimony came from the British officials themselves. They acknowledged, on the record, that the country’s AI oversight system is almost entirely voluntary — and that even a “too dangerous” verdict would not stop a release.

The UK AI Security Institute (AISI), which sits inside the Cabinet Office, tests frontier models before they launch — but only when companies voluntarily hand the models over. Henry de Zoete, the institute’s director, admitted that the institute depends entirely on the cooperation of companies and foreign governments to get access to models.

Pressed on what happens if AISI judges a model too dangerous and the company ships it anyway, de Zoete conceded the institute has no way to stop it. “We are not doing certification like a kite mark,” he said. “We don’t say yes or no to any model.” He also acknowledged the structural reality: the companies building the most advanced models are American or Chinese, which limits what Britain can do.

Cat Little, the Cabinet Office’s top civil servant, was asked whether the whole system is currently voluntary. “For the moment, yes,” she replied.

The access problem runs through Washington

The hearing surfaced a second vulnerability: Britain’s access to frontier models effectively depends on the US government.

Duff Gordon said OpenAI gives pre-release access to the American AI standards body and will continue giving it to the British AISI “for as long as we can do that under the US government.”

Booth-Smith acknowledged that Anthropic had initially been unable to give the British AISI full, unrestricted access to Mythos 5.1, a model released on September 1 for limited partner use. “That is a question for the US government,” he said, adding that the institute now has access. The backdrop, widely reported earlier this year, was a temporary US export control on Anthropic’s Mythos 5 and Fable 5 models in June, lifted at the end of that month.

In other words: a British safety agency’s ability to test models built by American companies depends on decisions made in Washington. For lawmakers trying to design an independent British oversight regime, that is an uncomfortable fact.

The whistleblowers’ warning

The researchers who spoke at the hearing were blunter than the companies. Daniel Kokotajlo, a former OpenAI researcher, said: “I don’t think these companies can be trusted to self-regulate.” Steven Adler, who left OpenAI in late 2024, said he had lost confidence in the industry’s safety culture. Geoffrey Irving argued that labs believe they are trapped in a race that governments cannot stop — “both are false,” he said. “They can simply stop.”

The hearing took place on the same morning that MI5 chief Ken McCallum publicly warned that AI could increase the threat from bioweapons — a reminder, for the committee, of what the stakes are.

What happens next

Britain currently has no AI law comparable to the EU’s AI Act. The government has favored targeted interventions — mandatory incident reporting and biosecurity measures — over a single sweeping bill, and ministers are reportedly preparing to introduce amendments to a cybersecurity bill this month that would mandate incident reporting and reviews.

The hearing did not change that plan directly. But it put something unusual on the parliamentary record: two of the companies that would be most affected by binding AI rules, asking for them — and two others, refusing. For a committee trying to show that voluntary pledges are not a substitute for law, it was exactly the testimony it wanted.

Source: Infobae — https://www.infobae.com/tecno/2026/10/13/openai-y-anthropic-le-pidieron-al-parlamento-britanico-que-los-regule-con-leyes-obligatorias/ ; Blocktempo — https://www.blocktempo.com/openai-anthropic-british-parliament-binding-ai-rules-deepmind-meta-decline/