What matters in AI.

Subscribe

UK MPs want a watchdog that can stop dangerous AI models

The recommendations are not binding, and at this time the UK government does not agree to a law.

A committee of UK MPs wants a new AI Safety and Security Bill. The MPs want the law to make a watchdog that can stop AI models while developers make them. All four firms that the committee heard (OpenAI, Google DeepMind, Anthropic and Meta) tell MPs that they do not release a model that they think is dangerous. They do not use the same yardstick for safety.

What is new

The watchdog can stop high-risk training and testing, stop a release, or stop an AI service if the risks are not in control. Developers of high-risk systems must give proof for their safety claims. External experts must have access to examine them.

The evidence

The report names ten holes in the UK rules for AI safety. Three are: no mandatory reports of incidents, small access for external testers, and weak protection for whistleblowers. The committee calculates that AI can add a maximum of £400 billion to the UK economy in ten years.

What is not known

The recommendations are not binding. The UK government does not agree to the law at this time. The AI Security Institute tests models at this time, but only when firms agree. It cannot stop a release. MPs want a different regulator to make firms obey.

Insight: Parliament writes the bill that the government does not offer

In October 2026, officials told MPs that the government does not agree to AI measures in the Cyber Security and Resilience Bill. Liam Byrne told them that a bill must make law. At this time the committee names a full bill. This seems a move to set the agenda before the government does. Monitor if ministers put AI in a bill in the next session.

Sources

  1. UK MPs want AI watchdog with power to pull the plug on dangerous modelstheregister.com
AI MATTER · NEWS · AI MATTER · NEWS ·18 NOV2026

Posted

Tags

Learn the terms in this story

Guides for this story

More in Policy

All Policy news