OpenAI and Anthropic Tell British Lawmakers: Regulate Us

WIRED reports that at a UK Business, Innovation, Science and Trade Committee hearing, Anthropic’s Liam Booth-Smith said safety rules should be binding and voluntary commitments are not enough. OpenAI’s Tom Duff Gordon backed a statutory framework and international convention, starting with mandatory incident reporting.

Google DeepMind’s Owen Larter emphasized standards instead of endorsing a statutory framework; Meta’s Rob Sherman pointed to existing regulators. The same day, MI5 director Ken McCallum warned about AI’s potential to change bioweapon stakes. UK AI Security Institute leaders told MPs their pre-release testing depends on company cooperation and they cannot stop a release they dislike.

Former lab staff including Daniel Kokotajlo argued companies cannot be trusted to self-regulate. Committee chair Liam Byrne said the hearing was meant to show voluntary self-regulation is unreliable and floated using Britain’s coming G20 presidency to push an international convention.

Two labs asking to be regulated is not altruism by default—it can freeze rules around practices they already tolerate and raise rivals’ costs. Even so, putting that ask on the British parliamentary record next to DeepMind and Meta’s reluctance is clarifying.

The AISI testimony is the gut punch. If the UK’s specialist institute cannot compel model access or block a dangerous release, “world-leading safety” is branding. McCallum’s bioweapons warning in one building and a voluntary regime admitted in another is exactly the contrast Byrne wanted televised.

We’re for binding incident reporting and independent evaluation with enough teeth to matter. We’re against theater conventions that leave Washington’s permissions as the real gate—as Anthropic’s restricted Mythos access episode illustrated. Middle powers that want leverage need something sharper than hope and voluntary handovers.

Listen to the whistleblowers without outsourcing policy to them. Listen to the labs without mistaking regulatory enthusiasm for surrender. Britain’s choice is whether AISI becomes a real auditor or a polite visitor.

The UK has leaned on voluntary cooperation between frontier labs and its AI Security Institute. Parliament is testing whether that model can survive as capabilities and geopolitical pressure rise.