OpenAI, Anthropic Back Binding AI Rules at UK Hearing — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- OpenAI and Anthropic told the UK Parliament's Business, Innovation, Science, and Trade Committee that voluntary AI safety pledges are insufficient, backing binding statutory frameworks — with OpenAI's Tom Duff Gordon calling mandatory incident reporting a starting point for any new law.
- Google DeepMind and Meta declined to back new legislation at the same hearing, with DeepMind's Owen Larter focusing on "standards" and Meta's Rob Sherman telling MPs they were "not writing on a blank page" because existing regulators already exist.
- Britain's AI Security Institute has no enforcement power — director Henry de Zoete acknowledged AISI cannot prevent companies from releasing models it judges too dangerous, and Cabinet Office permanent secretary Cat Little confirmed "at the moment, yes, it is" voluntary.
- MI5 director Ken McCallum, delivering his annual threat update blocks away from the hearing, warned that AI's potential to "change the stakes" on bioweapons is the misuse case he'd "attach greatest weight to" — a contrast committee chair Byrne called damning.
- Anthropic disclosed it was initially unable to provide full access to a frontier model to UK AISI due to US government restrictions, with Booth-Smith noting that "British access runs through Washington" before confirming AISI now has access.
- Whistleblowers Daniel Kokotajlo, Steven Adler, and Geoffrey Irving told MPs the industry cannot be trusted to self-regulate, with Irving calling the labs' belief in an unstoppable safety race "false" — they "could just stop."
- Committee chair Liam Byrne said he plans to use the UK's G20 presidency, followed by South Korea, to broker an international AI convention independent of the Trump White House — one that, he said, must outlive any single administration.
Why it matters: The hearing exposed a structural gap in Britain's AI oversight: AISI's director admitted the institute cannot block a dangerous model release, and Anthropic revealed it was once barred from sharing a frontier model with UK testers by the US government. With companies self-selecting into safety testing and British access running through Washington, the hearing pushed Parliament toward statutory reform and an international convention that doesn't depend on Trump's White House.
Ask SkimNews
