Why It Matters
Centralised AI can level the technological playing field for 43 forces while delivering measurable efficiency gains, but poor data or procurement missteps could undermine public safety and trust.
Key Takeaways
- •Police.AI centralises AI procurement for UK police forces
- •Live facial recognition vans to increase to 40 nationwide
- •Data quality remains critical bottleneck for AI effectiveness
- •£125 million budget contrasts with £1.1 billion NLEDS cost
- •Agile procurement needed to avoid costly delays
Pulse Analysis
The creation of Police.AI marks the most ambitious digital overhaul in British policing in two centuries, bundling AI research, procurement and policy under one roof. By standardising tools such as live facial‑recognition vans, deep‑fake detectors and predictive analytics, the Home Office hopes to accelerate adoption across the newly merged National Police Service. This centralisation promises economies of scale, faster innovation cycles, and a unified data strategy that could free millions of officer hours from routine paperwork, reshaping resource allocation in a sector under fiscal pressure.
However, the effectiveness of any AI system is only as strong as the data that fuels it. Legacy platforms like the Police National Computer have long suffered from fragmented records and inconsistent reporting, challenges that the upcoming NLEDS replacement aims to resolve. Yet the real obstacle may be human‑generated data quality; incomplete or erroneous entries can poison machine‑learning models, leading to false predictions or wrongful identifications. Elevating data‑maturity standards, revising crime‑recording guidance, and training staff on accurate entry are essential steps to ensure AI tools deliver reliable insights rather than amplifying existing biases.
Procurement strategy will be the make‑or‑break factor for Police.AI. Past centralised programmes, notably NLEDS, have spiralled in cost and schedule, underscoring the need for an agile, industry‑partnered approach. With a modest £125 million allocation, the hub must engage SMEs, enforce transparent tendering, and embed cybersecurity safeguards to protect public trust. Learning from the Ministry of Defence’s rapid acquisition reforms could help Police.AI avoid bureaucratic inertia, ensuring that AI deployments arrive on time, within budget, and with the public confidence required for sensitive policing applications.
Police.AI - New Tech Tools for UK Law Enforcement
By Elijah Glantz, Research Fellow, Organised Crime and Policing · Co‑Dr Pia Hüsch, Research Fellow, Cyber and Tech
With the National Centre for AI in Policing – dubbed Police.AI – UK policing is stepping up its pursuit of cutting‑edge tools, but cost‑effective, impactful delivery requires more than a new name.
Artificial Intelligence (AI) technologies for security purposes are widely associated with defence applications, from killer robots to drone targeting. Yet AI also offers a wide range of opportunities for law enforcement. As the pressure to analyse vast amounts of data increases for law enforcement officers amid a resource squeeze on policing, AI tools promise efficiency, speed and the hope to keep up with criminals.
The case for greater use of AI in policing is summarised by Sir Stephen Kavanagh, former Executive Director at INTERPOL:
“Criminal threats have moved on, and we haven’t. It is time for a new mindset: one that treats data and computer power as strategic assets.”
AI in policing is now set to receive a boost. The National Centre for AI in Policing, or Police.AI, was established amid a flurry of reforms announced by government in what the Home Secretary called “the most significant modernisation in nearly 200 years”. The freshly announced National Police Service (NPS) will merge several agencies, including Counter‑Terrorism Policing, the National Crime Agency and Regional Organised Crime Units. The exact architecture has yet to be decided, but the Home Office were clear on intentions to provide a single home for procurement, digital, data and technology policy in policing. This sets the stage for streamlined and expanded deployment of AI tools, nationwide.
The first example of centrally led technology deployment is the expansion of AI‑assisted live facial recognition (LFR) technology. Pilot programmes run by the Metropolitan Police Service have yielded impressive results and, owing to deliberate police communication campaigns, are very well received by the public. The Home Secretary announced the immediate scaling of the programme, delivering 40 more LFR vans across the country. Though catching headlines, LFR is just the tip of the iceberg; AI solutions range from predictive analytics to real‑time investigative support.
“With incomplete, irregular or incorrect data, the machine learning models risk identifying incorrect patterns, leading to wrong actions.”
Beyond facial recognition, the new Police.AI structure is set to invest in a range of AI technologies intended for rollout across law enforcement. These include advancing AI‑assisted deep‑fake detection, instant transcription translation and cutting‑edge digital forensics. Moreover, predictive analytics, or predictive policing, are a rapidly expanding AI use case already adopted by several forces. Research by the Police Foundation noted forces use predictive analytics for a range of tasks, including demand forecasting, domestic‑abuse risk management and developing intelligence to uncover complex links. Notably, the Ministry of Justice reportedly began work on a Homicide Prevention Project, euphemistically renamed “sharing data to improve risk assessment”. Now centrally coordinated, success cases of AI systems can be rolled out nationally.
With the multitude of AI tools under development and in use, the Home Secretary’s stated focus is to leverage AI to “free officers from paperwork”, claiming savings of up to 6 million hours. Similar government projections of millions of hours of police time saved have framed AI primarily as a resource‑saving tool. Real consideration must be given to leveraging AI to improve and acquire novel capabilities and improve performance. As argued regarding the wider police reform agenda, it is crucial to avoid “cost‑savings tunnel‑vision”.
Systems are Only as Good as Their Data
The drive to adopt effective and cost‑efficient AI and data‑led solutions is, at least partially, incompatible with the government’s wider drive to reduce police paperwork. Should efforts to slash bureaucratic requirements for officers lead to a reduction in data reporting and data quality, AI and big‑data‑driven police tools will be the first to suffer. Police use of AI will only be as good as the data it is trained on.
Unification and harmonisation of disparate datasets across UK policing is a daunting task. Law enforcement has long complained of issues and limitations in the National Computer (PNC) – operationalised in 1974. Issues reported range from poor user accessibility and unintuitive workflow to wholesale data wipes. The implementation of NLEDS, the PNC’s long‑awaited successor system, will look to remediate some of these challenges, improving data accessibility for law enforcement. Furthermore, the centralisation of policing will, in theory, streamline data sharing and harmonisation, breaking down long‑standing information siloes across agencies and forces. Though harmonisation of the disparate data sets across agencies and forces is a daunting task – as government as a whole has poor levels of “data maturity and governance”. The task of improving and harmonising systems notwithstanding, policing stands to benefit from a more accessible, centralised data pool from which to power its data‑driven tools.
As policing expands its use of AI tools, the quality of human‑input data may be a greater bottleneck than outdated, 1970s‑era computer systems. Many of the complaints around existing databases in fact relate to the quality of data – data reliant on human input. The risk is most pronounced in the development and deployment of predictive tools. Repeated studies on machine‑learning performance emphasise that accurate and high‑quality training data is foundational to accuracy.
When databases receive selectively recorded data, partially input records or otherwise incorrect inputs, it threatens to “poison” core training data. With incomplete, irregular or incorrect data, the machine learning models risk identifying incorrect patterns, leading to wrong actions. As a matter of public safety and trust, protecting against inadvertent – but dangerous – data poisoning must be a core priority. This requires deliberate thought in (re)designing the Home Office’s crime‑recording guidance and recording system in the forthcoming NLEDS, ensuring that data collected is done with a deliberate eye towards future analysis. Police staff – from front‑line cops to specialised investigators – must also be sensitised and aware of best practice and the strategic importance of robust data reporting. Together, this can improve police data “maturity” – improving the value‑add of the data collected for AI tools and wider policing.
Getting Procurement Right
One of the most compelling arguments for Police.AI is the ability to streamline procurement of AI systems for policing. Centralisation of procurement can offer better cost per unit and ensure that forces across the country have more equal access to tools. It can also prevent forces from pursuing duplicate programmes and ensure contractors are not left to navigate each force’s unique procurement systems, currently a risk as procurement plays out across England and Wales’s 43 forces, national bodies and specialist forces.
However, as HMICFRS’s 2025 Annual Assessment noted, “central isn’t always better; many national technology programmes have been overdue and over budget”. The National Law Enforcement Data Service (NLEDS) is case in point. Originally due to be operational in 2020 after a four‑year development period, the Home Office’s landmark project to modernise data use in policing has yet to become operational. The National Audit Office noted in 2021 that costs to the Home Office had already increased by 68 % to a total of £1.1 billion. The system is projected to come online in early 2026, almost six years behind schedule. Centralisation doesn’t eliminate risk but rather concentrates it. Putting procurement policy at the forefront of Police.AI is crucial to avoid replications of the lethargic and costly procurement process seen across government.
To this end, there are invaluable lessons to be learned from the MoD’s quest for a “faster and more agile” procurement system amid a rapidly changing threat environment. Police.AI’s procurement system should embrace early and frequent contact with industry, forming a policing‑industry partnership from the outset. Partnership forums, organised officially under Police.AI, should meet routinely to map policing needs and communicate advances in AI capabilities. This also includes ensuring small and medium enterprises can meaningfully compete in tenders – which is particularly important in the current AI development landscape, marked by fast‑moving start‑ups. Especially with the comparatively limited budget allocated to Police.AI of £125 million – considering NLEDS’ original £600 million price tag – maximising industry engagement, agility and delivery are paramount.
Finally, procurement must also consider key and evolving security considerations, such as dependency on dominant vendors, cyber‑security risks and geopolitical considerations, given these systems will be embedded into functions at the heart of UK public services. Retaining public trust is also key. The decision of some German police forces to procure AI technologies from the controversial US tech‑giant Palantir has illustrated how divisive the public debate can be on these issues.
Overall, Police.AI is a credible boost to the UK’s efforts to leverage AI technologies for both public services and national security. Yet for AI adoption in law enforcement to be successful, it needs more than one‑off investment and a new name. Human skills, data quality and procurement mechanisms need to adapt at scale and at pace to deliver on promised results.
0
Comments
Want to join the conversation?
Loading comments...