What Is the UK AI Safety and Security Bill?

The UK AI Safety and Security Bill is a proposed law that MPs want the government to introduce in the next session of Parliament. It would create a statutory regulator for the most powerful AI systems, with legal powers to delay, restrict or halt the training, testing and deployment of models judged too risky. It is not law yet: it is a recommendation from a House of Commons committee, and the government has not committed to introducing it.
The proposal comes from the Business, Innovation, Science and Trade Committee, which published its report, Guardrails for growth: AI safety and security, on 18 November 2026. The report is the eleventh report of the committee in the 2026-27 session, numbered HC 689. It follows months of hearings with AI researchers, former industry employees and senior representatives of OpenAI, Google DeepMind, Anthropic and Meta.
The committee's starting point is that Britain already has world-recognised expertise in testing AI systems, through the AI Security Institute, known as AISI, which was founded in 2023. But AISI's work depends on the voluntary cooperation of the companies whose models it tests, and it has no power to stop a company from releasing a model that raises safety concerns. The committee argues that technical expertise must now be matched by enforceable legal safeguards and a separate statutory regulator.
What would the bill do?
At the centre of the proposal is a new statutory regulator, distinct from the AI Security Institute. AISI would keep testing frontier models, while the new body would enforce the rules. The regulator would have legal powers to require independent verification of developers' safety claims, to pause high-risk training and testing, to delay or prohibit deployment, and to suspend services where risks cannot be controlled. Crucially, those powers would apply while systems are still being developed, not only after they have been released.
Developers of designated high-risk systems would have to provide evidence supporting their safety claims and give independent researchers meaningful access to examine their models before release. If a system's capabilities, safeguards or use changed materially, that would trigger fresh scrutiny, with restrictions or suspension where the risks could not be adequately controlled.
The committee also wants a clear legislative roadmap combining targeted action, duties to prevent harm, a clear liability regime, criminalising the misuse of AI for harm, and an international convention to enshrine global rules. The bill itself is described by the committee as merely the first step in that wider framework.
What are the ten gaps the MPs found?
The report identifies ten critical weaknesses in Britain's current approach to AI safety. The committee says ministers must address each of them. The gaps are: missing definitions of unacceptable harms and risk thresholds; a lack of statutory powers to access models; a lack of independent verification of developers' safety claims; insufficient controls on training, testing, internal deployment and containment; the absence of safety evidence before high-risk systems are released; the absence of mandatory reporting of incidents and near misses; a lack of powers to restrict or halt deployment; insufficient safeguards for models whose release cannot be reversed; inadequate liability and redress mechanisms; and insufficient protection for whistleblowers.
Several of those gaps go to a single problem: nobody has to tell anyone anything. At present there is no mandatory reporting of AI incidents or near misses, reporting arrangements are fragmented, and independent evaluators have no guaranteed access to models before release. The committee also heard that some models, once widely copied and distributed, cannot reliably be recalled, which is why it wants safeguards to begin during development rather than after deployment.
Former employees told the committee that raising safety concerns can carry a price. The hearings exposed commercial pressures, contractual restrictions and financial disincentives that can discourage staff from speaking out. The report therefore calls for protected channels for whistleblowers and stronger legal protections for them, alongside mandatory incident reporting.
Why do MPs say the bill is needed now?
The committee frames the case as economic as well as protective. Drawing on OECD productivity estimates, it calculates that AI could leave Britain's economy up to 400 billion pounds larger after ten years. It also argues that the UK's expertise in AI testing and evaluation could become a world-beating export industry in AI assurance, but only if the rules around it are credible.
Committee chair Liam Byrne MP said Britain could not give today's AI leaders unfettered freedom to behave like modern-day Dr Frankensteins, creating dangers beyond anyone's control. He said AI firms must be required to substantiate their safety claims, independent experts must be able to scrutinise them, and a regulator must have the power to intervene when risks cannot be controlled.
The evidence also showed a practical problem: the four companies that gave evidence all told MPs they would not release models they considered unsafe, but they do not use the same yardstick for safety, and they remain responsible for judging whether their own systems meet it. UK law does not generally require them to obtain an independent opinion before releasing anything. The committee's answer is to move safety from a promise into a duty.
What would change for AI companies?
For developers of designated high-risk systems, the changes would be substantial. They would have to evidence their safety claims, open their models to independent evaluators, and accept that significant capability changes trigger reassessment. They would also have to report incidents and near misses under a mandatory system, with standardised severity classifications and notification timelines, rather than the fragmented arrangements that exist today.
The committee also wants effective protections and routes to redress for individuals harmed by AI systems, clearer responsibility across AI supply chains, and criminal offences to deter harmful misuse and activities that enable it. Ministers, who are accountable to Parliament, would have to determine what risks society should accept, with AISI providing the technical evidence to inform those decisions.
For smaller businesses and ordinary users, the committee presents the bill as a confidence measure rather than a restriction. It argues that clear rules would give more businesses, especially small and medium-sized firms, the confidence to adopt AI, and give consumers support to use the technology safely and responsibly.
What happens next?
The committee wants the government to set out a legislative roadmap, including the AI Safety and Security Bill, in the next parliamentary session. It also wants ministers to publish, within three months, objectives for the UK's 2027 presidency of the G20, identifying negotiating priorities, responsible ministers and milestones, including cooperation on model access, safety assessments, incident reporting and high-risk releases, and an assessment of options for a multilateral convention or treaty.
Alongside the report, the committee launched a call for evidence on the UK's role in securing a global convention on AI safety and security through its 2027 G20 presidency. This is the first report in the committee's long-term inquiry into AI adoption, safety and governance, and it deals primarily with domestic measures needed in the short and medium term.
None of this is law yet. The committee's recommendations are not binding, and the government has yet to commit to the proposed legislation. Whether ministers will hand a watchdog the power to stop AI development, and how they will define which systems count as high-risk, is the decision the report is really asking for.
Frequently asked questions
Is the AI Safety and Security Bill law? No. It is a recommendation in a House of Commons committee report published on 18 November 2026. The government has to decide whether to introduce a bill; the committee wants one in the next session of Parliament.
Who would enforce it? A new statutory regulator, separate from the AI Security Institute. AISI would continue testing frontier models, while the regulator would enforce the rules and could pause training, block deployments or suspend services.
Does it ban AI? No. The proposal targets designated high-risk systems and gives a regulator powers to intervene where risks cannot be controlled. It does not propose a general ban on AI development or use.
What is the AI Security Institute? AISI, founded in 2023, is the UK's centre of expertise in AI testing, research and evaluation. It tests some of the world's most powerful models, but its work depends on voluntary cooperation and it has no enforcement powers.
When will we know if it becomes law? Only when the government responds to the report and sets out its legislative plans. The committee has asked for a roadmap including the bill in the next parliamentary session, but no date has been set.