Security News

UK MPs Call for AI Watchdog With Power to Halt Dangerous Models

The Register · 18 Nov 2026
Key Takeaway When choosing AI tools, ask suppliers what independent testing backs their safety claims, because a vendor's own assurance is currently the main safeguard.

A UK parliamentary committee is calling for a regulator with the power to intervene when AI models are judged too dangerous, rather than leaving that decision to the companies building them. In a report titled Guardrails for Growth: Making AI Safe, the Business, Innovation, Science and Trade Committee recommends an AI Safety and Security Bill. The watchdog could halt high-risk training and testing, block deployments, or suspend AI services if the risks cannot be contained.

The committee questioned researchers, former AI employees, and representatives from OpenAI, Google DeepMind, Anthropic and Meta. All four companies said they would not release models they considered unsafe. However, they do not use the same measure of safety, and UK law does not generally require an independent opinion before release. Committee chair Liam Byrne said that once some models are released, they cannot reliably be recalled.

The report identifies ten gaps in Britain's approach, including no mandatory incident reporting, limited access for independent testers, and weak protections for whistleblowers. Former AI employees told MPs that restrictive contracts and commercial pressure can discourage staff from raising concerns. The UK's AI Security Institute tests powerful models but relies on voluntary cooperation and cannot stop a release. MPs want it to keep testing while a separate regulator handles enforcement. Developers of designated high-risk systems would need to prove their safety claims and give independent researchers access.

The committee also argues that stronger safeguards could encourage businesses, particularly smaller firms, to adopt AI. It wants the laws passed in the next parliamentary session and hopes the UK will use its 2027 G20 presidency to push other countries towards similar rules. The recommendations are not binding, and the government has not yet committed to the legislation.

Why it matters

Businesses adopting AI services have largely assumed that vendors' internal safety checks are adequate, yet the committee heard that companies use different yardsticks and that UK law rarely requires independent review. The lack of mandatory incident reporting means customers may not learn about problems or near misses in the models they depend on. Until rules exist, the burden of judging vendor risk falls on the buyer.

What to do

  • Ask AI vendors what independent testing supports their safety claims.
  • Check supplier contracts for how and when they would notify you of AI incidents.
  • Track the proposed AI Safety and Security Bill, as its status is not yet committed by the UK government.
Answering to more than one framework? Run them as one security and AI governance programme ->

Summarised by CISO AI from The Register, written with Claude Sonnet 5.5. We link back to every original so you can read it yourself.