Executives from OpenAI and Anthropic told British lawmakers that voluntary safety commitments are no longer sufficient and called for binding legislation to govern frontier AI development, creating a public split with Google DeepMind and Meta, which refused to support new laws. The testimony came during a Tuesday evidence session held by the Business, Innovation, Science, and Trade Committee, a cross-party panel of MPs that heard from four leading AI companies, former employees turned whistleblowers, and senior British government officials. The hearing laid bare a fundamental disagreement among the world's most powerful AI firms over whether statutory rules are needed to ensure safety.
Liam Booth-Smith, Anthropic's head of Europe, told the committee that "the rules should be binding," saying his company has long held that voluntary pledges fall short and that mandatory requirements should apply to all developers. OpenAI vice president Tom Duff Gordon offered more cautious backing, supporting both a statutory framework and an international agreement while stating that "there should be some elements that should be mandatory," beginning with incident reporting. Google DeepMind's Owen Larter declined to endorse legislation when asked directly, instead saying his focus would remain on standards. Meta's Rob Sherman pointed to Britain's existing regulatory agencies, arguing that policymakers were "not writing on a blank page." The hearing also revealed significant limitations in Britain's current oversight system: Cat Little, the top civil servant at the Cabinet Office, confirmed under questioning that the entire regime is voluntary at present, while Henry de Zoete, director of the AI Security Institute, acknowledged his organization has no power to stop a company from releasing a model AISI deems too dangerous.
The schism between companies emerged as Britain's security service issued its own warning about AI risks. MI5 director Ken McCallum used his annual threat briefing, delivered just meters away from the parliamentary hearing, to identify bioweapon misuse as the most serious potential danger from artificial intelligence, saying AI could "change the stakes on existing or even new bioweapons." Committee chair Liam Byrne said after the session that his panel had set out "to light up the wide differences in opinions between some of the wealthiest firms on Earth" and demonstrate "that you couldn't rely on voluntary self-regulation." Former OpenAI researchers testifying as whistleblowers reinforced that message: Daniel Kokotajlo told MPs he doesn't believe the companies can be trusted to regulate themselves, while Geoffrey Irving said lab leaders think they're trapped in a race that governments won't stop and they can't exit—both assumptions he called false.
The hearing exposed Britain's dependence on Washington for access to cutting-edge models. Duff Gordon said OpenAI provides models before deployment to the US AI Safety Institute and will keep sharing with Britain's AISI "as long as we're able to do that under the US government," making clear that British access flows through American permission. Booth-Smith confirmed that Anthropic couldn't initially give unrestricted access to one model to UK officials, calling it "a question for the US government," though access was later granted. Byrne told WIRED his committee would now focus substantial effort on charting a path forward for Britain as a middle power on AI, hoping to use the country's upcoming G20 presidency alongside South Korea to negotiate an international convention independent of the current White House. "The G20 presidency will outlive the US presidency of Trump," he said, adding that "the risks are multiplying faster than our safety regime can control." The contrasting approaches of leading AI labs suggest that any international framework will face resistance from companies betting that voluntary standards can forestall harder rules, while rivals position themselves as allies of regulation they believe is coming regardless.