28 October 2026 · top ten
The JCHR's 'Human Rights and the Regulation of AI' Is Right About Everything. Here Are the Ten Parts That Won't Survive.
A person watching an AI decide something about you is not the same as a person in control. Parliament's Joint Committee on Human Rights said so in September1, in HC 160. The government has until mid-November to reply. Here are the ten findings it will quietly set aside.
10. The AI Bill promised in the 2024 King's Speech hasn't arrived. The JCHR is the fourth parliamentary body to ask for it. The government will reply by listing what it has done instead.
9. Risk tiers need to be mandatory, not suggested. Sector-by-sector guidance has no teeth. The JCHR asks for banned uses, pre-approval requirements, and due diligence across the whole supply chain. The government accepted all 44 recommendations of the NHS AI Commission on 6 October2, so it can say yes when the subject is healthcare. Whether it does the same when the subject is itself is a different matter.
8. The AI Security Institute needs legal powers. AISI runs on consent. Labs let it test their models when they choose to. The JCHR says put it on a statutory footing. Consent-based oversight stops when somebody decides to stop it.
7. High-risk AI needs a sign-off before it goes live. Not a review after something breaks. Approval first. The JCHR uses the drug comparison deliberately: you don't find out a drug kills people by prescribing it. By end of 2027, the government's reply to HC 160 will carry no commitment to pre-deployment sign-off for any AI system. If it does, I was wrong.
6. The whole supply chain should carry responsibility, not just the last link. A hospital buying an AI diagnostic tool didn't build it. If it fails, the hospital answers. The JCHR asks for due diligence obligations up the chain. The government will say existing law covers it.
5. You should know AI decided before you try to push back. The Public Law Project told the committee3 that people routinely can't challenge AI-assisted decisions because they don't know AI was involved. The JCHR calls this a transparency need. It is actually: tell people what decided about them.
4. Some uses of AI should be banned outright. The JCHR asks for prohibitions on uses that breach human rights. The UK government has refused to name prohibited things for three years running. Calling that flexibility is the word doing the most work in this debate.
3. An independent regulator with no engineers isn't independent. It's ignorant. The JCHR calls for a new statutory body to hold developers to account. A body that can't match frontier AI labs for staff or access doesn't regulate those labs. It reads their press releases.
2. A human watching an AI is not the same as a human deciding. JUSTICE told the committee4 that "even highly trained and qualified humans in the loop may struggle to scrutinise effectively the AI's output." The government's whole AI oversight strategy rests on a human reviewer being equivalent to a human decision-maker. The JCHR says they aren't. That finding will be noted and filed.
1. The harm lands hardest on the people who use public services most. The Public Law Project named this. The committee printed it. It is the only finding in 59 that says out loud who loses first. In 2031, the person this finding describes will still not know a scoring model processed her Universal Credit claim. The transparency requirement will still be coming. She is the one who won't appear in the reply by name.
Two months is long enough to accept every recommendation that costs nothing.