Parliamentary committee calls for legal duties on AI developers and a regulator able to stop high-risk systems reaching the public.

MPs and peers have called for a dedicated UK AI law that would make developers responsible for risks built into their systems and give a new regulator the power to prevent unsafe systems from being released. In a report published today (14 September), the Joint Committee on Human Rights says existing laws leave gaps in protection against discrimination, invasions of privacy and decisions that people cannot effectively challenge.

High-risk systems would require approval before they could be provided or used, while organisations would also have to tell people when AI has contributed to an important decision about them and explain its role.

The problems

The committee’s principal objection is that UK law tends to place responsibility on the organisation using an AI system, even when the problem was built into it by its developer. A company might, for example, use a recruitment tool that disadvantages certain applicants without knowing enough about its training data or design to identify the cause.

The report cites evidence of an Uber Eats driver unable to log on because facial recognition software failed to identify him. It also describes workers flagged for investigation after stopping at traffic lights while driving – an example of a system treating routine behaviour as grounds for suspicion. The committee also heard about AI used to profile prisoners and scan people’s faces in public places.

Existing laws offer some protection. Data protection rules govern the use of personal information, while equality law can provide a remedy for discrimination. But the committee says those laws do not consistently require developers to identify and tackle human rights risks before their systems reach customers. It also warns that powerful suppliers may pass responsibility through their contracts to smaller businesses that lack the information or resources to investigate a problem.

“The major gap in our regulatory system is managing risk upstream with those building the tech,” Michael Birtwistle of the Ada Lovelace Institute told the inquiry.

The proposals

The committee wants a law covering those who design, develop, supply and use AI. Their duties would depend on their role and the risk posed by the system. Lower-risk tools would face lighter requirements, while high-risk systems would need approval before they came to market.

Some uses would be banned entirely. The committee identifies subliminal manipulation, emotional inference and inappropriate use of profiling or biometric data as possible targets for prohibition. It also recommends barring the development and provision of exceptionally powerful systems, including artificial general intelligence or superintelligence, if they could cause widespread and very serious harm, including by evading effective human control. The committee says the scope of any bans should be settled after public consultation.

The proposed bill would also require organisations to say when AI is used in decisions that have serious consequences for people and explain what role it played. Moreover, the committee wants to strengthen the safeguards that apply before such decisions are made. Under the UK GDPR, involving a person in the process can prevent a decision from being treated as solely automated, but the committee says that protection is meaningful only if the person understands the AI’s output and is free to question it.

An independent oversight body would investigate complaints, set enforceable codes of practice and maintain a public record of AI incidents. It would be able to test high-risk systems, prevent their release, order their withdrawal and require remedies for people harmed by them. Developers of powerful AI models would also have to submit them to the AI Security Institute for testing. At present, the institute relies on developers agreeing to give it access.

The committee is not alone in calling for a law covering AI across the economy. Conservative peer Lord Holmes of Richmond, who introduced an AI regulation bill in 2025, told a Lords debate in June that job applicants could be rejected without knowing AI had played a part or having a practical way to challenge the decision. Questioning the government’s reliance on existing regulators, he asked what would happen in a sector with “no competent regulator”. That bill did not progress beyond first reading.

The government’s position

Ministers have so far relied on existing regulators. In a written answer on 10 September, AI minister Kanishka Narayan said most AI systems were “already regulated at the point of use by our existing expert regulators”, adding that the government was keeping those arrangements under review.

But committee chair Alex Sobel MP says regulation at the point of use cannot adequately deal with risks already built into a system. “New legislation is needed to establish a comprehensive set of protections that deal with the entire AI supply chain and its lifecycle,” he said in the committee’s announcement.

GLI

Message History

Delivery option(s)