Twelve MPs and peers from three parties have told the UK government that its approach to artificial intelligence is not fit for purpose, that no other country's is either, and that the fix is a law Westminster promised in July 2024 and has never introduced. The government now has roughly two months to say whether it agrees.
The Joint Committee on Human Rights published its report, "Human Rights and the Regulation of AI", on September 14. It runs to about 100 pages, makes 20 recommendations, and draws on ten oral evidence sessions and more than 70 written submissions. Its conclusion is blunt: the UK's patchwork of sector-specific laws applies mostly at the point of deployment, leaves regulators unable to test or block a system before release, and makes it hard for anyone harmed by an AI decision to find out it happened, let alone challenge it.
"Nowhere in the world, including the UK, has a current legislative and regulatory approach to AI that is fit for purpose," said committee chair Alex Sobel, a Labour MP. "New legislation is needed to establish a comprehensive set of protections that deal with the entire AI supply chain and its lifecycle. A single AI regulator should be established to set policy, monitor performance and with the teeth to ensure enforcement."
The committee wants a dedicated AI Bill built around a risk-based regime. Low-risk systems would face light obligations; higher-risk systems heavier ones, some requiring prior approval before deployment. Certain uses would be banned outright, with subliminal techniques and inappropriate profiling or biometric processing named as candidates. Due diligence duties would apply to every actor in the supply chain, graded by role and risk, and people would have to be told when AI is used in decisions that significantly affect them.
At the centre sits a new institution. "A single, independent AI oversight body should be established on a statutory basis," the report says. "The body would act as the central point of contact for raising concerns about the use of AI and carry out oversight and monitoring of AI harms and risks." It would set codes of practice, keep a public register of AI incidents, and hold powers to sanction, to pull systems from the market, and to grant remedies to individuals. The committee suggests it could be built by expanding the AI Security Institute, which tests frontier models only when developers volunteer them and has no statutory power. That point has fresh bite: the Financial Times reported this month that Anthropic declined to submit its newest model to AISI, and AISI had nothing to impose in return.
The report records who liked the status quo. Meta, it says, "warmly welcomed" the government's approach. Human rights, legal and technical witnesses called it "uncritical and deregulatory" and "asleep at the wheel".
The government's timeline invites that criticism. The July 2024 King's Speech pledged to "seek to establish the appropriate legislation to place requirements on those working to develop the most powerful artificial intelligence models." No bill followed. This year's King's Speech contained no AI bill; ministers pointed to deepfake measures, cyber security legislation and AISI's testing. On September 1 the government rejected Lord Clement-Jones's narrower proposal for a legal power to shut down a dangerously behaving frontier model, with the Cabinet Office telling the BBC that Britain "cannot simply turn AI off". A spokesperson responding to the JCHR said any future measures would be "evidence-based".
Industry's response was the surprise. The same day, OpenAI's head of European policy, Tom Duff Gordon, said the company backs "stronger UK rules for the handful of companies, including OpenAI, developing the most powerful AI systems" and urged ministers to "take advantage of this political window which is clearly opening up". The scope matters: OpenAI wants obligations confined to frontier labs and to national security and cyber risks. That is a far narrower bill than the JCHR's, touching a handful of incumbents rather than the thousands of UK firms deploying AI in hiring, credit, policing and benefits, which is where the committee says harm is already occurring.
Three days later, King Charles hosted around 30 executives and ethicists, including Nvidia's Jensen Huang and Google DeepMind's Demis Hassabis, at Dumfries House to discuss principles for frontier development. The King stayed 20 minutes. No commitments emerged.
Why It Matters
The JCHR has done what the government has avoided for two years: written down what a UK AI law would contain. Nicola Cain of the compliance consultancy Handley Gill, whose evidence the committee cited, called the proposal "a watered-down version of the EU's AI Act", and the comparison holds. Prohibited practices, prior approval for high-risk systems and lifecycle duties map onto the EU regulation's Articles 5, 43 and 27. Where the JCHR goes further is redress: a regulator able to grant remedies to individuals is something the EU Act does not offer, and Cain called it "revolutionary".
The politics are awkward. Labour's 2024 manifesto promised binding rules for "the handful of companies developing the most powerful AI models", and OpenAI is now offering ministers that wording back with industry endorsement attached. Accepting it would let the government claim delivery while leaving the deployment-stage harms the JCHR documented untouched. The report's one soft spot, a recommended ban on developing artificial superintelligence that echoes a private member's bill Sobel tabled the week before, would be symbolic given the relevant labs are American, and critics will use it to dismiss the rest. They should not: most of the report concerns discriminatory automated decisions and biometric scanning happening in the UK now.
What to Watch
Government responses to select committee reports are expected within two months, putting the deadline in mid-November. The test is whether it engages with the 20 recommendations or restates the line that AI is best regulated at the point of use by existing regulators. Watch for whether ministers fold OpenAI's frontier-only proposal into a narrow bill, whether AISI gets statutory power to compel model access after the Anthropic episode, and whether the Council of Europe AI treaty the UK signed but has not ratified becomes the hook for something broader. The committee has named the gap. Filling it remains, for now, optional.
"Nowhere in the world, including the UK, has a current legislative and regulatory approach to AI that is fit for purpose."— Alex Sobel MP, Chair, Joint Committee on Human Rights