UK human rights committee asks for an AI Bill and mandatory model testing by AISI

Lords and MPs ask for an AI Bill on 14 September

On 14 September 2026 the UK Parliament's Joint Committee on Human Rights published its fourth report of the 2026-27 session. The report is titled "Human Rights and the Regulation of AI". It carries the numbers HC 160 and HL Paper 56. The committee's publication record lists release at 00:01 UK time, which was 07:01 SGT.

The committee makes three central asks. The government should introduce the AI Bill it promised in the 2024 King's Speech (para 178). It should place the AI Security Institute (AISI) on a statutory basis (para 231). Developers of powerful models should have to submit new models to AISI for testing (para 232).

The inquiry began in July 2025. The committee received over 70 pieces of written evidence and held ten oral evidence sessions. Alex Sobel MP chairs the committee. Lord Alton of Liverpool took the chair at the 9 September meeting that agreed the report.

The committee's press release quotes Sobel. "Nowhere in the world, including the UK, has a current legislative and regulatory approach to AI that is fit for purpose," he said.

No legal force and a two-month reply window

A committee report has no legal force. It asks the government to act. The report page says the government has two months to respond. By our count, that points to a reply in mid-November 2026.

The report describes a changed Whitehall map. The Department for Science, Innovation and Technology has been abolished, and its functions have moved elsewhere (para 78). AISI now sits in the Cabinet Office (para 76). Kanishka Narayan is Minister of State for AI, jointly in the Cabinet Office and the new Department for Business, Innovation, Science and Trade (para 78).

Today, labs give AISI access to their models on a voluntary basis. "The AISI has no statutory power," the committee concludes (para 109). The report adds that AISI cannot demand access to foundation models before or after deployment (para 106).

Mandatory submission and a power to withdraw

The new element is compulsion. Para 232 says developers "must be required to submit new models (and new versions of existing models) to AISI". The stated purpose is "review, evaluation, and testing". They would also have to hand over technical specifications. The list covers model properties, training data and process, intended use, security and safety testing, and risk controls.

AISI would gain a duty to publish findings before release. It would also issue preliminary "warning" statements on new foundation models (para 231). If AISI finds significant risks to human rights, it would advise a new oversight body to block a launch or order a withdrawal (para 233).

That oversight body is the second big ask. Para 226 calls for "an independent AI oversight body on a statutory footing". Para 227 lists its powers. One power is "to prohibit AI models or systems from being launched or deployed, and to order their withdrawal from the market". The body would also keep a public repository of AI incidents. It could limit or prohibit UK deployment by developers based outside the UK.

The Bill itself would follow a risk-based and precautionary model. The committee draws an analogy with pharmaceuticals (para 186). High-risk systems would need prior approval before deployment (para 204). Para 200 goes further. It would prohibit very powerful systems "(such as Artificial General Intelligence and Artificial Superintelligence)" that risk "the capacity to evade effective human control".

The committee cites the OpenAI and Hugging Face incident, which OpenAI described on 21 July 2026 (para 140). It concludes that society "largely relies on private companies to take voluntary action" against such incidents. "This is unsatisfactory," the committee says (para 143).

Thresholds and timing stay open

The report sets no compute threshold or capability test for a "powerful" model. It gives no date for a Bill. The committee says it did not try to prescribe the exact details of a framework (para 176). It leaves the detail of any prohibitions to a public consultation (para 201).

The minister's evidence points the other way. He told the committee that AISI "is unparalleled globally in its access to models pre-deployment" (para 229). He said he saw no need for reform to its access, and he promised to keep the question under review. The government has said before that it would put AISI on a statutory footing. It has not done so (para 229).

As of 28 September, the committee's publications list shows no government response. A GOV.UK search also found none.

Map the para 232 dossier before November

• Policy teams at model developers can watch for the government response in mid-November. Its answer on para 232 decides whether pre-release submission moves toward law.
• Developers can map the dossier that para 232 describes against their current model cards and system cards.
• Providers outside the UK can read para 227. The proposed body could limit or prohibit their UK deployments.
• Compliance teams can read paras 207 and 208. They name the EU AI Act as "a useful point of departure" for due-diligence duties.
• Teams with no UK exposure can file the report for reference. It changes no legal duty today.