The Home Office promised police AI would roll out with strong oversight and a public registry of every tool in use. Nine months on, the oversight consultation's response is overdue, and the registry still has no rules for what it will show.
Estimated reading time: 6 minutes
In short. The Home Office's January 2026 Police Reform White Paper promised that AI rolled out to police forces in England and Wales would be "supported by a new regulatory framework with strong oversight and accountability" [1], resting on a consultation whose own text promised "a response to this consultation exercise is due to be published within 12 weeks" of closing [2]. The consultation closed on 12 February 2026 [3]; no response has been published, roughly 24 weeks past that deadline. The same White Paper named a public registry of every AI tool in police use as one of three founding functions of the new national centre, PoliceAI, due to deliver "from Spring 2026" [1]. The Home Office's own factsheet, updated yesterday, says the policy on what that registry will disclose "is currently being developed", with only a pilot of its public-facing element promised "by the end of 2026" [4]. Two documented mistakes: four stars.
Police forces in England and Wales are already using artificial intelligence at scale. Since 2024 the Home Office has put over £50m of direct grant funding into AI projects, including live facial recognition vans now deployed by seven forces, and the Police Reform White Paper commits a further £115m over three years [4] [1]. A new national centre, PoliceAI, hosted by the College of Policing, mobilised in April 2026 and formally launched in June [4]. The Home Office's own factsheet on police use of AI, last updated 22 October 2026, sets out what the government says is in place to govern all of this. Checked against the White Paper the factsheet implements and the consultation it points to, two of the governance commitments that were meant to accompany the rollout have slipped well behind the pace of the rollout itself.
What was promised, and when
The White Paper's own words tie deployment and oversight together explicitly: "we will give the police the resources and expertise to deploy AI in an ethical, robust and responsible way, supported by a new regulatory framework with strong oversight and accountability. Through Police.AI we will also create a public facing registry of the AI being deployed by police forces and the steps they have taken to ensure the reliability of tools before being used for operations" (page 13) [1]. Later in the same document, the registry is listed as one of three core functions PoliceAI was built to deliver from the outset: "a strategy, oversight and coordination function will provide public facing communications on police AI use, including publishing and maintaining a registry of the AI tools the police are using and the steps they've taken to test and evaluate them before deployment" (page 86), alongside the commitment that "we will set up Police.AI to begin delivering from Spring 2026" (page 86) [1].
The "new regulatory framework" rests on a specific consultation. The Home Office launched it on 4 December 2025, covering the legal basis for live, retrospective and operator-initiated facial recognition and similar technologies, and it closed on 12 February 2026 [3]. The consultation document itself sets its own deadline for what comes next: "A response to this consultation exercise is due to be published within 12 weeks on GOV.UK" [2]. Twelve weeks from 12 February is 7 May 2026.
The claims, tested
| The document's own words | What the record shows | Verdict |
|---|---|---|
| "supported by a new regulatory framework with strong oversight and accountability", tied explicitly to the AI rollout (White Paper, page 13) [1] | The framework's first building block, the facial recognition consultation, promised its own response "within 12 weeks" of its 12 February close [2]. Checked directly on the live consultation page today, no response is listed and the page's own modification date still reads 4 December 2025, the day it opened [3] | The response is roughly 24 weeks overdue by the department's own deadline, with deployment continuing regardless |
| PoliceAI's 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" (factsheet) [4] | The same answer says "a policy on what should be disclosed is currently being developed and agreed with policing leaders", with only "a pilot of the public-facing element" targeted "by the end of 2026" [4] | A founding function promised from PoliceAI's "Spring 2026" delivery date still has no agreed disclosure rules five months after the centre's formal launch |
| "The government expects all Chief Constables to complete the registry once it is launched and will act where this is not the case" [4] | No launch date for the full registry is given anywhere in the factsheet, only the pilot target for its public-facing element; what counts as "act[ing]" where a force fails to complete it is not defined | An enforcement promise with no defined trigger, since the registry it depends on has no confirmed launch date |
The mistakes, counted
The oversight framework the rollout was promised alongside is roughly half a year behind its own deadline, with no acknowledgement of that in the department's current public material (1). The White Paper ties "strong oversight and accountability" directly to the AI deployment it is funding, in the same sentence as the registry commitment [1]. The facial recognition consultation that was meant to build that framework set itself a 12-week response deadline, now roughly 24 weeks overdue [2] [3]. The factsheet's own facial recognition answer, current as of yesterday, still describes the consultation only as something the government "launched... to develop" a framework, with no mention that its promised response is overdue [4]. Meanwhile the facial recognition deployment the framework is meant to govern has not paused: the White Paper commits to extending live facial recognition vans to more forces in the same document [1].
A founding transparency function of PoliceAI has no agreed disclosure rules five months after launch (2). The White Paper named the public registry as one of three core functions PoliceAI exists to deliver, from a "Spring 2026" start [1]. PoliceAI formally launched in June 2026 [4]. As of the factsheet's 22 October 2026 update, what the registry will actually disclose is still being worked out, and the only firm date offered is a pilot of its public-facing element by the end of the year, not the registry itself [4]. Public scrutiny of what AI tools individual forces are running, the specific mechanism the White Paper offered as the public's check on the programme, remains undeliverable in the meantime.
Credit where due
The factsheet does not hide any of this. It states plainly, in its own words, that the registry's disclosure policy is still being developed and gives an honest pilot date rather than an invented one, and it says clearly that no findings have been reached on the facial recognition framework rather than implying progress that has not happened [4]. PoliceAI's own account of what it will not do, including ruling out hosting operational AI models itself and ruling out making ethical judgements in place of Chief Officers, is specific and testable rather than vague reassurance [4]. This piece makes no finding about the competence or conduct of any named official; the criticism is of the pace of the department's own published commitments against the pace of the deployment they are meant to govern.
Verdict
Two mistakes, both about sequencing rather than any false statement: the oversight framework is running roughly half a year behind the deadline it set for itself, and the transparency registry that was supposed to accompany PoliceAI from day one still has no rules for what it will show, five months after launch. Four stars. None of this stops AI deployment in policing, which is proceeding on schedule or ahead of it; it is the governance half of the bargain that has fallen behind.
The star score counts two documented mistakes: the facial recognition consultation's own 12-week response deadline is roughly 24 weeks overdue with no acknowledgement in the department's current public material, and PoliceAI's public registry, one of its three founding functions from a promised "Spring 2026" start, still has no agreed disclosure policy five months after formal launch. One fall in the 1 to 3 band: four stars; the bands are on the ratings page. Checked directly against the Police Reform White Paper (January 2026), the live facial recognition consultation page and its accessible consultation document, and the Home Office's police AI factsheet, all fetched 23 October 2026.
Sources
- "From Local to National: A New Model for Policing", Home Office Police Reform White Paper, CP 1489, January 2026
- "Consultation on a new legal framework for law enforcement use of biometrics, facial recognition and similar technologies (accessible)", Home Office, published 4 December 2025
- "Legal framework for using facial recognition in law enforcement", GOV.UK consultation page, Home Office, opened 4 December 2025, closed 12 February 2026
- "Police use of artificial intelligence (AI): factsheet", Home Office, published 9 June 2026, last updated 22 October 2026