Skip to main content
AI & Machine Learning

UK MPs Push for AI Watchdog with Power to Halt Risky Models

Formal UK parliamentary setting with table, chairs, and podium, filled with natural light and papers, conveying a serious…

"We cannot give today’s AI leaders unfettered freedom to behave like modern-day Dr Frankensteins, creating dangers beyond anyone’s control," said committee chair Liam Byrne MP.

Parliament's Business, Innovation, Science and Trade Committee recommends an AI Safety and Security Bill

The committee has published a report, Guardrails for Growth: Making AI Safe, that calls for primary legislation to create an AI watchdog with statutory powers. The new bill would shift Britain away from relying solely on developers' assurances and toward a regulator able to act before risky systems are released.

Powers the committee wants: halt training, block deployments, suspend services

The report would give a regulator authority to intervene while models are still being developed — including the ability to halt high-risk training and testing, to block deployments, or to suspend AI services where risks cannot be contained. The committee emphasises that large language models, trained on billions of parameters, can behave unpredictably and “once some models are released, they cannot reliably be recalled,” as Byrne warned.

Ten safety gaps identified, including incident reporting, independent testing and whistleblower protections

The committee lists ten holes in the UK's current approach. The report highlights three particulars: the lack of mandatory incident reporting, limited access for independent testers, and inadequate protections for whistleblowers. MPs found that those gaps combine with legal and commercial incentives that can deter frontline staff from raising concerns.

What MPs heard from companies, researchers and former employees

During hearings the committee questioned researchers, former AI employees, and representatives from OpenAI, Google DeepMind, Anthropic, and Meta. All four companies told MPs they would not release models they considered unsafe — but the committee notes those firms do not use the same yardstick for safety, and UK law does not generally require them to obtain an independent opinion before releasing models. Former employees told MPs that restrictive contracts, commercial pressures, and financial disincentives can discourage staff from speaking out about safety problems.

The AI Security Institute (AISI): testing expertise, voluntary reach

The UK already hosts an AI Security Institute (AISI), which "has made a name for itself testing some of the world's most powerful models." MPs recommend AISI continue that independent testing work but say enforcement should sit with a separate regulator. AISI's work currently depends on voluntary cooperation and it lacks authority to prevent a company releasing a model that raises safety concerns, the report notes.

How developers, regulators, and whistleblowers are affected

  • Developers and vendors: The committee would require makers of designated high-risk systems to provide evidence supporting safety claims, to allow independent researchers access to examine models, and to submit significant capability or safeguard changes to further scrutiny.
  • Regulators and independent testers: MPs want a statutory regulator with the power to intervene during development and deployment, while keeping AISI as a testing body whose findings could feed enforcement action.
  • Whistleblowers and staff: The committee calls for stronger legal protections and for compulsory reporting of AI incidents and near misses, responding to testimony that current contracts and commercial pressures can penalise staff who raise safety concerns.

The committee also framed the ask with an economic argument: using OECD productivity estimates, it calculates AI could add up to £400 billion to Britain's economy over the next decade. MPs see stronger safeguards as a means to encourage adoption by businesses — particularly smaller firms — and as a commercial opportunity for the UK to monetise independent model testing expertise.

MPs want the new laws in the next parliamentary session and say Britain should use its G20 presidency in 2027 to press other countries to adopt tougher rules on AI testing, incident reporting, and risky model releases. The report's recommendations, however, are not binding: the committee acknowledges that ministers must decide whether to give a watchdog the power to stop AI development, and the government has yet to commit to the proposed legislation.

The core choice laid out by the committee is clear: continue to depend on private companies’ varying safety standards and voluntary cooperation, or legislate a regulator with the power to intervene before potentially dangerous models are put into the wild. Whether ministers will hand a watchdog the authority to “pull the plug” remains the next, concrete political decision.

Original report