
An AI Safety and Security Bill would give Britain a regulator with the power to halt dangerous AI models, under proposals published on 18 November 2026 by MPs on the Business, Innovation, Science and Trade Committee. The recommendations are not binding, and ministers have yet to commit to legislation.
The committee’s report, Guardrails for Growth, argues that Britain is leaving critical safety decisions to the companies building the most powerful systems. After hearings with researchers, former industry staff and senior figures from OpenAI, Google DeepMind, Anthropic and Meta, MPs identified ten gaps in UK safeguards, according to the committee’s own summary.
What would the AI Safety and Security Bill do?
The committee wants the bill in the next parliamentary session. It would sit within a wider legislative roadmap that includes duties to prevent harm, a clear liability regime and criminal offences for using AI to cause harm.
Central to the plan is a new statutory regulator, kept separate from the AI Security Institute (AISI). The report says that body should have legal powers to require independent verification of safety claims, pause high-risk training and testing, delay or prohibit deployment, and suspend services where risks cannot be controlled. AISI would carry on supplying technical evidence, while ministers, accountable to Parliament, would decide which risks society accepts.
The AISI’s work currently relies on voluntary cooperation. As The Register reported, it has no authority to stop a company releasing a model that raises safety concerns.
The ten gaps MPs found in UK AI safety
The committee’s list includes missing definitions of unacceptable harms and risk thresholds, no statutory powers to access models, and no independent verification of developers’ safety claims. It also flags the absence of mandatory safety assessments before high-risk systems are deployed, no mandatory reporting of incidents and near misses, and no powers to restrict or halt unsafe development.
The remaining gaps cover models whose release cannot be reversed, weak liability arrangements and routes to redress, and insufficient whistleblower protection. Developers told MPs they would not release unsafe models. But the committee found that companies apply different safety thresholds and remain responsible for judging whether their own systems pass.
What did former AI insiders tell MPs?
Evidence quoted by Computer Weekly was blunt. Daniel Kokotajlo, a former OpenAI governance researcher, said he left because he lost confidence in the company’s leadership to behave responsibly. Steve Adler, another former OpenAI employee, described pressure not to use words such as “dangerous” and to play up benefits rather than risks.
Liam Byrne, the committee’s chair, said the country cannot give AI leaders “unfettered freedom to behave like modern-day Dr Frankensteins”. He added: “Once some models are released, they cannot reliably be recalled. Protection must begin while systems are being built and tested.”
Why the £400 billion figure matters
The report is framed as pro-growth rather than anti-AI. Drawing on OECD productivity estimates, the committee calculates that AI could leave Britain’s economy up to £400 billion larger after ten years. It also sees a chance to build a world-beating export industry in AI assurance, using the testing expertise developed at AISI.
Clear rules, it argues, would also give small and micro-businesses the confidence to adopt the technology. The report further warns that increasingly open-weight models from Chinese developers are cheap, easy to adopt and hard to control once released.
What it means for you
Nothing changes today. The report is a set of recommendations to government, which has two months to respond, and there is no bill yet. If one is introduced, it could eventually mean mandatory incident reporting from AI firms, stronger protection for staff who raise concerns, and clearer routes to redress if an AI system harms you.
For businesses using AI tools, the direction of travel is towards more accountability for developers and clearer supply-chain responsibilities. The committee also wants ministers to publish, within three months, objectives for the UK’s 2027 G20 presidency, so that international cooperation on model access and incident reporting runs alongside action at home. It has opened a call for evidence on securing a global AI safety convention.
What happens next
The timing is notable. Anthropic’s blockbuster flotation, covered in our guide to buying Anthropic shares in the UK, has put frontier AI developers firmly in the public eye, and recent disruption such as the Microsoft outage has sharpened questions about dependence on a few technology suppliers. AI is also now finding serious software flaws, as with Java4Shell.
Whether ministers will hand a regulator the power to stop AI development is the open question. The government’s response is due within two months. Follow the latest in our technology coverage.

































