Sunlit Houses of Parliament in London showcasing neo-gothic architecture

UK MPs call for an AI Safety and Security Bill and a regulator that can halt frontier models

By Yash Malviya·

Photo: AXP Photography / Pexels

A Commons committee report on November 18, 2026 lists ten gaps in UK AI safety law and wants a new regulator with power to pause training. The government has not committed, and the prime minister is pitching a G20 deal instead.

A cross-party group of MPs wants Britain to write frontier AI safety into law. On November 18, 2026, the Commons Business, Innovation, Science and Trade Committee published "Guardrails for Growth: Making AI Safe," which calls for an AI Safety and Security Bill in the next parliamentary session and a new regulator with the power to stop work it judges too risky.

The report lands two days after the prime minister told the Lord Mayor's Banquet that Britain would seek "a new consensus" on AI through the G20. The two approaches pull in different directions: a binding domestic regime now, or an international compact later.

What the committee found

The committee took evidence from AI researchers, former industry staff and senior representatives of Anthropic, Google DeepMind, Meta and OpenAI. According to The Register's report, all four companies said they would not release models they considered unsafe, yet they share no common safety standard and UK law generally does not require independent review before release. Resultsense's analysis adds that each company sets its own thresholds and decides whether its own systems pass.

The report names ten gaps. They include no defined thresholds for unacceptable harm, no statutory power to access models, no independent check on developers' safety claims, weak controls over training, testing and internal deployment, no mandatory reporting of incidents or near misses, no power to restrict or halt unsafe development, and weak whistleblower protection. Ex-industry witnesses described commercial pressure and contract terms that discourage speaking up.

The regulator it wants

The design separates three jobs. Ministers, accountable to Parliament, would decide how much risk society accepts. The UK AI Security Institute (AISI) would keep supplying technical evidence. A new regulator, not AISI, would enforce.

“We cannot give today's AI leaders unfettered freedom to behave like modern-day Dr Frankensteins.”

Liam Byrne MP, committee chair, as quoted by The Register, November 18, 2026

The proposed powers are broad. According to Resultsense, the regulator could order independent checks on developers' safety claims, halt training and testing judged high-risk, hold back or block a deployment, and suspend services whose risks cannot be controlled. Developers of high-risk systems would have to substantiate safety claims and give independent evaluators meaningful access, and any material change to a system's capabilities or safeguards would trigger a fresh assessment.

Committee chair Liam Byrne is quoted by The Register as saying that "Once some models are released, they cannot reliably be recalled," and that "Protection must begin while systems are being built and tested." That second point is the real shift, since today's debate is mostly about what happens after release. The committee also wants mandatory incident reporting, protected whistleblower channels and clearer liability.

Iconic view of Big Ben and Westminster Palace in London on a sunny day
The Palace of Westminster, where the Commons committee's report will now go to government for a reply. Photo: Matheus Bertelli / Pexels

The money argument

The report frames safety as a growth condition. Using OECD productivity estimates, the committee says AI could make Britain's economy up to £400 billion larger over the next decade, and argues that businesses, especially smaller ones, need credible safeguards before they adopt it with confidence. It also sees a chance for the UK to earn revenue from independent model testing. That framing is deliberate: it answers ministers who fear regulation will push frontier developers elsewhere.

Where the government stands

The report is advisory. No formal government response appeared in the coverage we reviewed, and the government has not committed to the bill. Resultsense notes that a search of the bills register for "artificial intelligence" returns four bills, all private members' bills, none an Act. Our UK AI regulation tracker follows the same gap between pressure and legislation, and our report on the Cyber Security and Resilience Bill shows the narrower route ministers have preferred.

On November 16, Prime Minister Andy Burnham said the UK would use its G20 presidency to "seek a new consensus on perhaps the most contentious issue of our times: AI." Citing the AI Security Institute, he added that "Britain is well-placed to lead the world towards new global guardrails" at next November's leaders' summit, the "Manchester Compact on AI." The committee's report is clear, per Resultsense, that chasing an international deal must not delay rules at home, and it wants ministers to set out their G20 objectives within three months.

“Britain is well-placed to lead the world towards new global guardrails”

Andy Burnham, Prime Minister, Lord Mayor's Banquet speech, November 16, 2026

How the committee got here

This is the second parliamentary push in two months. On September 14, the Joint Committee on Human Rights published "Human Rights and the Regulation of AI," which The Next Web reported also wanted a dedicated AI law and an independent oversight body. That report noted the AI Security Institute tests models before release but has no regulatory authority and cannot demand access. On the same day, OpenAI's head of policy for Europe, the Middle East and Africa, Tom Duff Gordon, told Politico that OpenAI could support legislation setting "durable, mandatory, capability-based requirements for frontier AI," limited to the few labs at the frontier.

The same coverage reported that Anthropic had skipped UK pre-release tests for Mythos 5.1, according to the Financial Times, and that Business Secretary Jonathan Reynolds warned people not to get "hyperbolic" about AI risks, arguing that regulation cutting Britain off from frontier developments "would obviously make us less safe." That tension, between access to models and power over them, is the heart of the new report.

Our take

The report is a strong diagnosis and a thin forecast. The ten gaps are real: AISI's access rests on goodwill, and no outside body can check a lab's own safety claim. But a committee cannot legislate. The signal to watch is the government's reply and whether an AI bill appears in the next King's Speech. If it does not, the Manchester Compact becomes a substitute for law instead of a complement to it, and the UK's testing regime stays voluntary for another parliamentary year.

Frequently asked questions

What did the UK MPs' AI safety report recommend?

The Business, Innovation, Science and Trade Committee called for an AI Safety and Security Bill in the next parliamentary session and a statutory regulator separate from the AI Security Institute, with power to pause high-risk training, delay or block deployment and suspend services.

What is the name of the November 2026 UK committee report on AI?

It is titled "Guardrails for Growth: Making AI Safe" and was published by the Commons Business, Innovation, Science and Trade Committee on November 18, 2026. It identifies ten gaps in UK AI safeguards.

Will the UK government pass an AI safety bill?

Not yet. The committee's recommendations are not binding and the government has not committed to the proposed bill. Resultsense reports the four AI bills on the register are all private members' bills, and none has become law.

What is the Manchester Compact on AI?

It is the name Prime Minister Andy Burnham gave on November 16, 2026 to the UK's goal for the 2027 G20 leaders' summit: new global guardrails on AI. The speech did not set out what the compact would contain.

Would the regulator replace the AI Security Institute?

No. The committee wants AISI to keep testing models and supplying technical evidence. A separate regulator would enforce the rules, and ministers would decide how much risk is acceptable.

Why does the committee say the UK needs AI safeguards?

It says voluntary cooperation underpins AISI's work, companies set their own safety thresholds, and released models cannot reliably be recalled. It also argues, citing OECD estimates of up to £400 billion in gains, that credible safeguards help adoption.

Sources

What each one is, and whose it is.

  1. OtherThe vendor’s own
  2. Press reportIndependent of the vendor
  3. Press reportIndependent of the vendor
  4. Press reportIndependent of the vendor
  5. Press reportIndependent of the vendor