UK MPs push for AI watchdog that can switch off risky models
A UK parliamentary committee wants a new law letting a regulator halt high-risk AI training runs, block deployments and suspend services while models are still being built. MPs say OpenAI, Google DeepMind, Anthropic and Meta can't be left to mark their own homework. The report lands Wednesday, and ministers haven't signed up yet.
- Original source: theregister.com.
- The source published it on Nov 18, 2026 at 9:30 AM.
- No second outlet has reported it yet. Treat it as a single-source claim.
The Business, Innovation, Science and Trade Committee is calling for an AI Safety and Security Bill that would let a watchdog step in while models are still in the lab. According to a report shared with The Register before Wednesday's publication, that regulator could halt high-risk training and testing, block deployments, or suspend AI services outright when the risks can't be contained. The document is titled Guardrails for Growth: Making AI Safe.
MPs took evidence from researchers, former AI employees and representatives of OpenAI, Google DeepMind, Anthropic and Meta. All four companies told the committee they wouldn't release a model they judged unsafe, but they don't apply the same safety yardstick, and UK law generally doesn't require an independent opinion before launch. The report names ten gaps in Britain's approach, including no mandatory incident reporting, limited access for outside testers and thin protections for whistleblowers.
Committee chair Liam Byrne MP didn't soften it: 'We cannot give today's AI leaders unfettered freedom to behave like modern-day Dr Frankensteins, creating dangers beyond anyone's control.' He said the hearings exposed holes in independent scrutiny, incident reporting and powers to stop unsafe systems, warning that 'once some models are released, they cannot reliably be recalled.' Protection, he argued, has to start while systems are being built and tested.
Britain already runs an AI Security Institute, which has tested some of the world's most powerful models, but it relies on voluntary cooperation and can't stop a release. MPs want the AISI to keep testing while a separate regulator owns enforcement. Developers of designated high-risk systems would have to produce evidence backing their safety claims and let independent researchers examine the models, with significant changes to capabilities or safeguards triggering fresh scrutiny. Byrne said firms 'must be required to substantiate their safety claims' and a regulator must be able to intervene when risks can't be controlled.
Former employees told MPs that speaking up carries a price, citing restrictive contracts, commercial pressure and financial disincentives. The committee wants stronger legal cover for whistleblowers plus compulsory reporting of AI incidents and near misses. There's money in it too: using OECD productivity estimates, MPs calculate AI could leave the UK economy up to £400 billion larger over the next decade, and they see an opening for Britain to sell independent model testing, building on the AISI's expertise.
The recommendations aren't binding, and the government has yet to commit to the legislation. MPs want the laws on the books in the next parliamentary session and want Britain to use its turn at the G20 helm in 2027 to push other countries toward tougher rules on AI testing, incident reporting and risky model releases. On the final call, Byrne was blunt: 'Ministers, accountable to Parliament, must decide what risks the public should be asked to bear.'
TopicsOpenAI · Google DeepMind · Anthropic · Meta · Liam Byrne · AI Security Institute · Business, Innovation, Science and Trade Committee · OECD
Source: theregister.com
Written by AI from verified reporting. Not financial advice.
More stories
- Tech & AI x.com 1 viewJensen Huang Retains Right to Name Two Board Directors Via 3.5% Stake, Reuters Says
- Tech & AI cryptobriefing.com 1 viewSpaceX reportedly leasing $47B in Nvidia chips via Apollo-backed vehicle
- Tech & AI coindesk.com Caught by @gimbung 1 viewGoogle finds an AI that hunts and exploits software flaws without a human
- Tech & AI blockchain.news 2 viewsXingchen unveils Xing 4, a 285B-parameter model built on Ascend chips

