OpenAI and Anthropic Want Britain to Regulate AI. Google and Meta Don't.

Four of the world's biggest AI companies sat before British lawmakers this week and couldn't agree on the most basic question in AI policy: should anyone be forced to follow the rules, or is asking nicely enough?

At a Tuesday evidence session held by Parliament's Business, Innovation, Science, and Trade Committee, representatives from OpenAI and Anthropic told the cross-party panel of MPs that voluntary safety pledges are no longer sufficient and that binding legislation is needed. Google DeepMind and Meta disagreed, pointing instead to existing regulators and voluntary standards. The split, aired on the parliamentary record in front of four former lab employees turned whistleblowers and senior UK civil servants, laid bare just how far apart the frontier AI industry remains on whether it should be forced to answer to anyone but itself.

Anthropic's head of Europe, Liam Booth-Smith, was unambiguous. "The rules should be binding," he told the committee, adding that the company has "consistently believed that there should be binding rules on companies and that voluntary commitments are not enough." OpenAI's vice president Tom Duff Gordon struck a more cautious note but landed in the same place, backing a statutory framework and an international convention. He argued incident reporting should be the starting point for mandatory rules, saying: "We don't think basing everything on voluntary commitments for frontier risk is a good long-term plan."

Their counterparts at Google DeepMind and Meta wouldn't go there. Asked point-blank whether he supported a statutory framework, DeepMind's Owen Larter demurred: "Focus on the standards is what we would be doing." Meta's Rob Sherman leaned on Britain's regulatory status quo, telling MPs the UK was "not writing on a blank page."

A deliberate exposure

According to committee chair Liam Byrne, surfacing this divide was the whole point of the exercise. He told WIRED the committee set out "to light up the wide differences in opinions between some of the wealthiest firms on Earth" and to demonstrate "that you couldn't rely on voluntary self-regulation." Putting the clash on record in a second major jurisdiction, after similar fault lines have emerged elsewhere, gives British lawmakers something concrete to point to as they weigh whether to legislate.

The timing made the stakes harder to ignore. While the four executives were testifying, MI5 director Ken McCallum was delivering his annual threat update just a few hundred meters away, and he flagged AI as a growing concern for national security. "Not yet realized, but the misuse case to which I'd attach greatest weight, is AI's potential to change the stakes on existing or even new bioweapons," McCallum said.

Byrne saw the overlap as more than coincidence. "The fact that the head of MI5 is warning about incredibly dangerous capability uplifts for bioterrorists in one part of Whitehall, and then in another part of Whitehall the permanent secretary of the Cabinet Office is basically admitting that our regulatory regime for AI is voluntary, tells you everything you need to know," he told WIRED.

The safety net has no teeth

That admission came in the hearing's final stretch, when civil servants were asked directly how much the current system actually compels. The answer: not much. Britain's AI Security Institute, known as AISI, tests frontier models before they're released to the public, but only when the companies that built them agree to hand them over.

Henry de Zoete, AISI's director, acknowledged the institute's access depends entirely on voluntary cooperation from both AI companies and foreign governments. Cat Little, the Cabinet Office's top civil servant, was asked outright whether the whole regime was voluntary. "At the moment, yes, it is," she said.

It got more striking from there. Asked what would happen if AISI flagged a model as too dangerous but the company released it anyway, de Zoete conceded the institute has no power to stop that. "What we're not doing is kite marking," he said, using the British term for a certification stamp. "We're not saying yes or no to any sort of model." He added a blunt assessment of Britain's actual leverage: "These companies are either American or Chinese who are building at the frontier... There's only so much we, as the UK, can do in that kind of context."

The companies themselves confirmed that Washington, not London, is the real gatekeeper. Duff Gordon said OpenAI provides its models to the US AI Security Institute, known as CAISI, before deployment, and will keep sharing with the UK's AISI "as long as we're able to do that under the US government." Anthropic's Booth-Smith was even more direct about the friction this creates. He told the committee the company had not been able to give the UK's AISI full, unrestricted access to a model referred to as Mythos 5.1 at the time of its release. "That's a question for the US government," he said, before noting that access has since been granted.

Former insiders, blunter words

The committee also heard from four former employees of frontier labs, whose testimony was sharper than anything offered by the sitting executives. Daniel Kokotajlo, who left OpenAI, told MPs plainly: "I don't think the companies can be trusted to self-regulate." Steven Adler, who departed OpenAI in late 2024, said he was "not confident in the safety culture across the industry." Geoffrey Irving pushed back on the industry's favorite justification for moving fast, the idea that labs are locked in a race no government will interrupt and none of them can afford to leave. "Both of those are false," he said. "They could just stop."

What comes next

Byrne told WIRED the committee will now spend significant time mapping out a path forward for the UK as what he called a "middle power" on AI policy, one without the leverage of Washington or Beijing but with some room to shape the conversation. He wants ministers to use Britain's upcoming G20 presidency, followed by South Korea's turn in the chair, to push for an international convention that doesn't hinge on the current occupant of the White House. "The G20 presidency will outlive the US presidency of Trump," he said.

His closing assessment of the hearing was stark: "The risks are multiplying faster than our safety regime can control."

For now, that regime remains exactly what AISI's own director described it as: a system built on companies choosing to cooperate, with no mechanism to stop them if they choose not to. Whether Parliament moves to change that, and whether OpenAI and Anthropic's public support for binding rules survives contact with actual legislation, is the question the committee says it plans to keep pressing.