AI Policy & Society // POLICING

The Home Office's £141m AI Policing Package Keeps the Ethics with Chief Officers

The Home Office set out new laws for police facial recognition this week. Its own factsheet, updated the day after, says the new national AI centre will not judge whether a tool should be used.

The Home Office said on 4 November 2026 that it will legislate to govern police use of facial recognition and create an independent watchdog which, in extreme cases, could ask a court to stop a force using the technology. The same department's guidance on police use of artificial intelligence, updated the following day, describes the machinery already being assembled behind it: a national centre that sits inside a GBP 115 million White Paper programme for police AI and automation, and an unusually explicit list of things it will not do.

According to the factsheet, the government supported police use of AI with more than GBP 50 million of investment in 2024-25 and 2025-26. The Police Reform White Paper announced a further GBP 115 million for police adoption of AI and automation, which the document says covers a range of projects including the creation of the centre, and GBP 26 million more is going to a national facial recognition system, taking the overall White Paper investment package to GBP 141 million. The GBP 115 million also pays for 40 new live facial recognition units. Separately, the department's 4 November statement said an GBP 11 million investment is tripling the number of live facial recognition vans available to forces in England and Wales.

The new National Centre for AI in Policing, known as PoliceAI, began mobilising in April 2026 and was formally launched in June. It is hosted by the College of Policing on behalf of forces in England and Wales, funded through a grant agreement issued by the Home Office under Section 57 of the Police Act 1996, and is intended to move into the National Police Service once that body is established. Its director, the factsheet says, is Alex Murray, previously the National Police Chiefs' Council's AI lead, and the Home Office says it did not partner directly with any private company to deliver the centre; the College of Policing separately appointed Deloitte as lead delivery partner after a procurement exercise.

In its first year the centre is to concentrate on case file and disclosure assistants, crime data integrity tools, rapid analysis of CCTV and digital media, and image identification and classification. It will administer an AI Lab that tests, benchmarks and tunes commercial and open-source models in a sandbox before forces deploy them, offer forces funding to cover set-up, compute and training costs, and collate guidance. It also hosts the Policing AI Threat Hub, which the factsheet says works on detecting AI-generated deepfakes and on protecting the criminal justice system from AI-generated false evidence.

It will also maintain a public registry of the AI models police use. According to the factsheet, the registry “will set out what each AI tool in use by the police does, why it is used, the risks identified and how these risks are mitigated, as well as technical details”. The document adds that the government expects all Chief Constables to complete the registry once it is launched and will act where this is not the case. A pilot of the public-facing element is intended by the end of 2026.

The limits are stated as plainly as the ambitions. The centre “will not make ethical judgements on whether an AI tool should be used”, the document says; nor will it take operational decisions on the output of models. It will not host models for day-to-day operational policing, will not “undertake cyber and information security assessments of AI models” (the Police Digital Service holds that role), and will build new models from scratch only by exception, where a use case is unique to policing and cannot be met by open-source or purchasable ones. Those responsibilities stay with Chief Officers, who the factsheet says are accountable to locally elected Police and Crime Commissioners “until the positions are abolished and replaced”.

The facial recognition announcement follows a Home Office consultation opened in December 2025 on a new legal framework for biometrics and facial recognition. Under the plan a new watchdog would advise on standards, scrutinise police use and recommend to ministers on emerging technologies; where forces fail to comply it could pursue enforcement action leading to a court injunction. The statement says any technologies used by law enforcement must be rigorously and independently tested against bias or discrimination of age, gender or race, and that a specific law will make clear that police use of facial recognition must be intelligence led and proportionate.

The supporting figures are the Home Office's own. It says static live facial recognition cameras piloted in Croydon between October 2025 and March 2026 led to 173 arrests, and that the Metropolitan Police reported a 10.5 per cent drop in overall crime in the area against the previous year and a 21 per cent decrease in offences linked to violence against women and girls. At this year's Notting Hill Carnival, it says, police using the technology stopped 86 registered sex offenders, leading to seven arrests for breach of civil orders or notification requirements. It also says the Home Office has independently tested a new retrospective facial recognition algorithm found to be highly accurate and without bias, which forces will be able to roll out from next year.

The Home Secretary, Shabana Mahmood, described facial recognition as the biggest breakthrough for catching criminals since DNA matching, saying it had already helped police take thousands of dangerous criminals off the streets. She said its use must be fair, safe and proportionate, and that clear laws were needed so the public could have confidence the tools are used responsibly.

FOR HUMAN CONSUMPTION

Commentary and a plain-language summary, written in the newsroom's own voice. It is not reporting, it introduces no source, and it should not be quoted as a finding.

The police in England and Wales are getting more artificial intelligence. This week the government explained how it will be watched.

On 4 November it said it will write new laws about facial recognition, the technology that scans a crowd and tries to match faces to a list of wanted people. A new independent watchdog would keep an eye on it. If a police force broke the rules badly enough, the watchdog could go to court to stop that force using the technology.

The day after, the government updated a long document about police AI. It explains a new national centre, called PoliceAI, which will test AI tools before police use them and keep a public list of what is being used, so people can see. The centre is one of several projects paid for out of GBP 115 million for police AI and automation set out in the government's policing reform plan. A further GBP 26 million is going on a national facial recognition system.

The same document says what the centre will not do. It will not decide whether using a particular AI tool is the right thing to do. It will not check whether the tools are safe from hackers. Those jobs stay with the police chiefs in charge of each force, who answer to people who are elected locally.

That is the part worth thinking about. A centre full of experts can test whether a machine gets the right answer. Whether the machine should be pointed at people at all is still a decision made by a human being, and that human being is the one you can vote out.

The Humans Are Fine.

THE HUMANS ARE FINE.

Sources

Home Office, "Police use of artificial intelligence (AI): factsheet", first published 9 June 2026, updated 5 November 2026 Opened and read in full this run. Source of the GBP 50 million, GBP 115 million, GBP 26 million, GBP 141 million and 40-unit figures, the PoliceAI remit and limits, the registry description and the quotations in the article attributed to the factsheet. It states no GBP 11 million figure.
Home Office news story, "Stronger laws for police using facial recognition technology", published 4 November 2026 Opened and read in full this run. Source of the watchdog and legislative proposals, the GBP 11 million live facial recognition vans investment, the Croydon and Notting Hill Carnival figures, the bias-testing wording and the Home Secretary's quotation. It names the PoliceAI director as Al Murray; the factsheet spells the name Alex Murray.

Fact-check notes

Verified against the primary source: The 4 November 2026 announcement of new facial recognition laws and an independent watchdog. The GBP 50 million, GBP 115 million, GBP 26 million, GBP 141 million and 40-unit figures, read in the factsheet; no one of these is a budget for the centre itself, and the article no longer says so. The GBP 11 million live facial recognition vans figure, read in the 4 November statement; the factsheet contains no GBP 11 million figure. PoliceAI's launch timetable, host body, funding mechanism and director. The centre's stated limitations, including that it will not make ethical judgements on tool use and will not perform security assessments. The plan for a public registry and the government's stated expectation that Chief Constables complete it.
Attributed, not independently confirmed: All crime-reduction and arrest figures, which are the Home Office's account of police reporting. The claim that a new retrospective algorithm was independently tested and found highly accurate and without bias: asserted by the department, with no published results. The Home Secretary's characterisation of facial recognition as the biggest breakthrough since DNA matching.
Unknown: What an AI Lab sandbox will actually test, in what conditions, and whether its results will be published. Which models police forces currently use: the registry that would answer this does not yet exist publicly. The identity and powers of the proposed facial recognition watchdog, which exists as an announcement rather than a published bill in the retrieved material. The methodology behind the Croydon crime-reduction percentages and the bias testing of the retrospective algorithm. When the National Police Service will be established, which determines when PoliceAI's hosting arrangements change.
Editorial note: Both sources are Home Office publications. No independent reporting on either document could be retrieved this run, so the article reports government statements as government statements. The factsheet is standing guidance first published in June 2026 and updated on 5 November 2026; it is not presented here as a new document, and only the fact of the November update is relied on. The article does not describe the watchdog as existing: the retrieved statement announces an intention to create one. GBP is spelled out in the body because the house template escapes markup and currency symbols are not relied on. Revision 2026-11-05: an earlier version of this article described GBP 115 million as the centre's own budget line and GBP 141 million as the centre's plan. Neither figure is a budget for the centre: GBP 115 million is the White Paper allocation across several projects, of which the centre is one, and GBP 141 million is the whole White Paper package. The headline, lede, body and summary have been corrected, and the GBP 11 million figure has been reattributed to the 4 November statement, which is the only retrieved document that states it.

Evidence file: police-ai-centre-accountability-evidence.html