In short: A Commons committee says voluntary safety pledges are not enough and wants an AI Safety and Security Bill plus a regulator separate from the AI Security Institute. Published 21 Nov 2026 by the AI Decoded desk; news dated 18 Nov 2026.

A House of Commons committee has called for a new AI regulator with legal powers to step in while powerful AI models are still being built, rather than leaving safety decisions to the companies making them.

What the report says

The Business, Innovation, Science and Trade Committee published Guardrails for growth: AI safety and security on 18 November 2026. After hearing from researchers, former AI staff, and OpenAI, Google DeepMind, Anthropic and Meta, MPs identified ten gaps in the UK's safeguards.

They want an AI Safety and Security Bill in the next parliamentary session, and a statutory regulator that could require independent checks of safety claims, pause high-risk training and testing, delay or block a launch, and suspend services where risks can't be controlled.

Why now

All four companies told MPs they wouldn't release a model they considered unsafe. The committee's point is that they don't use the same yardstick, and UK law doesn't require an independent opinion before release. The AI Security Institute (AISI) also told MPs a model can't be recalled once its weights are copied and shared.

Where AISI fits

AISI would keep doing technical testing, but the committee says it "is not a regulator". The new body would be separate, drawing on AISI's expertise, with multi-year funding. The Government has two months to respond.

The UK angle

MPs frame this as good for business: they estimate AI could make the UK economy up to £400 billion larger over ten years, and say clear rules would help more firms adopt it. That figure is a scenario, not a forecast.

FAQs

What is the AI Safety and Security Bill?

A proposed law, recommended by MPs, that would cover high-risk AI training, testing and deployment and require reporting of serious AI incidents. The Government has not yet agreed to it.

Would the AI Security Institute become the regulator?

No. The committee wants a separate statutory regulator that draws on AISI expertise, while AISI continues its technical testing.

Sources

Follow AI Decoded on Substack

Weekly plain-English AI news and verified UK events.