UK MPs are pushing for a formal AI safety watchdog with the power to halt dangerous AI models during development.
What MPs Are Asking
A parliamentary committee has recommended an AI Safety and Security Bill.
The bill would create a regulator with authority to stop high-risk training, block model deployments, or suspend AI services if safety risks can’t be controlled.
Intervention must happen early—during testing and development—because once a model is live, it’s hard to recall or fix.
Why the Power to ‘Pull the Plug’ Matters
Large language models are trained on billions of parameters.
This allows them to learn complex patterns but also increases the risk of harmful, biased, or misleading outputs.
Developers can’t predict every possible response, especially in real-world interactions.
Once live, AI models scale across platforms and amplify errors, making oversight critical.
Key Findings from the Committee
The committee held hearings with AI researchers, former engineers, and executives from OpenAI, Google DeepMind, Anthropic, and Meta.
All four companies said they wouldn’t release unsafe models.
But each defines safety differently—based on internal rules, business goals, or technical thresholds.
UK law doesn’t require independent safety reviews before any AI product is released.
Concerns Raised by Former AI Staff
Former employees said raising safety concerns often leads to professional retaliation.
They face restrictive contracts, being sidelined, or job loss.
One DeepMind employee reported a safety warning was dismissed and followed by a transfer to a less visible role.
MPs now want legal protections for whistleblowers, including immunity from dismissal and mandatory incident reporting.
Existing Oversight and Its Limits
The UK has the AI Security Institute (AISI), which tests powerful AI models.
But AISI works voluntarily. It has no power to stop a company from releasing a model.
Its findings are not binding. Companies don’t have to share data or allow full access.

Proposed Regulatory Changes
MPs want AISI to keep testing, but with a new, enforceable regulator.
Developers of high-risk AI systems—like those in healthcare or finance—must prove safety with detailed documentation.
Independent researchers must be allowed to examine these models under strict rules.
Any major change to a model’s design or safeguards triggers new scrutiny.
Why This Matters for the UK Economy
Using OECD data, the committee estimates AI could add up to £400 billion to the UK economy over the next decade.
Strong safety rules could build public trust and encourage both big and small firms to adopt AI.
The UK could also earn revenue by offering independent AI testing through AISI.
What the Government Has Said
The committee’s recommendations are not binding.
The government has not committed to the proposed law.
Ministers have not said whether they will let a watchdog stop AI development.
Next Steps
MPs plan to introduce the AI Safety and Security Bill in the next parliamentary session.
They also aim to use the UK’s G20 leadership in 2027 to push global standards for AI safety.
Parliament’s Business, Innovation, Science and Trade Committee chair, Liam Byrne MP, said: ‘We cannot give today’s AI leaders unfettered freedom to behave like modern-day Dr Frankensteins, creating dangers beyond anyone’s control.’
Byrne added: ‘Once some models are released, they cannot reliably be recalled. Protection must begin while systems are being built and tested.’
The report identifies ten gaps in the UK’s AI safety framework, including:
- Lack of mandatory incident reporting
- Limited access for independent testers
- Inadequate protections for whistleblowers
- Insufficient transparency in model training data
- Weak enforcement mechanisms for safety claims
- Failure to address bias and discrimination in AI outputs
- No standardized definition of ‘high-risk’ AI systems
- Insufficient public consultation in AI policy design
- Underdeveloped post-deployment monitoring
- Lack of international coordination on AI safety
MPs also want stronger legal protections for employees who report safety issues. Without them, internal pressure could silence voices and reduce innovation.
For now, the UK remains one of the few countries without a formal, enforceable AI safety regulator. The proposed legislation is a key step toward a safer, more accountable AI ecosystem.
Sources & further reading
Featured image: original publisher image. Via original publisher. Image source



