OpenAI and Anthropic Back Binding UK AI Regulation as Google and Meta Hold Out

What happened

At a hearing of the House of Commons Business, Innovation, Science and Trade Committee, representatives for OpenAI and Anthropic told British lawmakers something tech companies rarely say: please regulate us. Both backed binding UK AI regulation, arguing that voluntary safety pledges are no longer enough. Google DeepMind and Meta, testifying at the same session, declined to support new laws.

Anthropic's head of Europe, Liam Booth-Smith, was the most direct, saying voluntary commitments are not enough. OpenAI's Tom Duff Gordon backed a statutory framework and an international convention, starting with mandatory incident reporting. DeepMind's Owen Larter pointed to standards work instead of laws, and Meta's Rob Sherman pointed to the UK's existing regulators.

Four former lab researchers also testified and were blunt: OpenAI alumnus Daniel Kokotajlo said the companies cannot be trusted to regulate themselves, and former staffer Steven Adler said he is not confident in the safety culture across the industry. On the same day, MI5 director Ken McCallum warned in his annual threat update that AI could raise the stakes on bioweapons — the misuse case he says he takes most seriously.

The UK's limited power

Officials were candid about how little authority the current system carries. The AI Security Institute tests frontier models before release only when companies choose to share them. Its director admitted the institute depends entirely on company cooperation and could not stop a company from releasing a model it judged too dangerous. Anthropic also acknowledged it could not initially give UK testers full access to its Mythos 5.1 model — a question for the US government, not the lab.

What this means for your business

- Mandatory incident reporting is coming. Both OpenAI and Anthropic signaled support for it, making it the likely first step in any new law. Start building AI incident logging and reporting into your AI deployments now — retrofitting compliance later is expensive. - Procurement questions are getting sharper. If you buy AI from frontier labs, expect enterprise customers and regulators to ask what safety evaluations your vendors have passed. Vendor questionnaires will evolve from security checklists to AI governance checklists. - Plan for regulatory divergence. The US runs on voluntary accords while the UK moves toward binding law. Multinationals should architect AI governance that can meet the strictest jurisdiction they operate in, rather than rebuilding for each one. - Transparency is becoming table stakes. Anthropic's published research on topics like distillation defenses may become mandatory disclosure rather than voluntary. Build your internal AI documentation as if regulators will eventually read it — because they might.

What to watch

Whether the committee's push produces an actual UK AI bill, and whether the UK uses its coming G20 presidency to advance an international AI convention. The question is shifting from whether AI gets regulated to how fast.