Wednesday, 18 November 2026 Tech news and buying guides for the US, UK, Canada and Australia AboutContactRSS
RoxMag
AI

UK MPs Back AI Watchdog With Power to Halt Dangerous Models

A parliamentary committee wants a regulator able to stop high-risk AI training and block releases, saying firms should not police themselves.

UK MPs Back AI Watchdog With Power to Halt Dangerous Models

UK lawmakers want a new regulator that could stop dangerous AI systems in their tracks, according to The Register. A parliamentary committee has recommended an AI Safety and Security Bill giving a watchdog the authority to act while models are still being built, not only after they ship.

The report, seen by The Register before its Wednesday publication, comes from the group of MPs that questioned researchers, former AI staff and representatives of OpenAI, Google DeepMind, Anthropic and Meta. Under the proposals, the watchdog could halt high-risk training and testing, prevent a model from being deployed, or suspend a service when its risks cannot be contained.

The committee argues that developers of large language models cannot fully predict how their systems will behave, because such models are trained on billions of learned parameters. All four companies told MPs they would not release a model they judged unsafe. However, the committee found they apply different standards, and UK law generally does not oblige them to seek an outside opinion before launch.

Liam Byrne MP, who chairs the committee, said today's AI leaders should not be given unrestricted freedom to create dangers nobody can control. He added that the hearings revealed weaknesses in independent scrutiny, incident reporting and the ability to stop unsafe systems, and that some models cannot be reliably recalled once released. Protection, he said, has to start during development and testing.

The report, Guardrails for Growth: Making AI Safe, lists ten gaps in Britain's current approach. These include no mandatory incident reporting, restricted access for independent testers and weak safeguards for whistleblowers. Former AI employees told MPs that restrictive contracts, commercial pressure and financial penalties can deter staff from raising concerns.

The UK already hosts an AI Security Institute, which tests powerful models, but its work relies on voluntary cooperation and it cannot block a release. MPs want it to keep testing while a separate regulator handles enforcement. Firms building designated high-risk systems would need to back up safety claims with evidence and let independent researchers examine their models, with major changes in capability or safeguards triggering fresh checks.

The committee also points to economic stakes, estimating that AI could add up to £400 billion to the UK economy over the next decade, based on OECD productivity figures. It sees scope for the UK to earn from independent model testing and argues that clearer safeguards could help smaller businesses adopt the technology.

Based on reporting by The Register. RoxMag adds plain-English context and checks every figure against that source.

AI regulationUK ParliamentAI safetyOpenAIDeepMind

More in AI