Artificial intelligence has been edging its way into police work for nearly two decades, from predicting crime hotspots to scanning hours of body-cam footage. But the arrival of powerful generative AI tools such as ChatGPT has supercharged the conversation - and raised urgent questions. How can police forces use these tools responsibly? How do they avoid bias, maintain transparency, and keep the public on their side?
A new article in Policing: A Journal of Policy and Practice introduces the Transformer Led Policing (TLP) model, the first structured framework designed specifically to help law enforcement integrate generative AI into their daily work.
A model built for policing
The TLP framework, developed by researcher Eric Halford, breaks AI adoption into three cyclical phases: Devise, Discuss, and Deploy. Each phase contains four elements, ranging from defining real-world use cases and designing prompts to testing, evaluating, and openly discussing issues of legality, ethics, and security. Unlike ad hoc approaches that might bolt AI onto existing systems, the TLP model is meant to ensure that every stage of adoption is deliberate, transparent, and auditable.
This structure matters because police organizations have historically been cautious about external technologies. Officers are more likely to adopt tools that feel tailored to their work rather than imported wholesale from other industries. By creating a framework explicitly designed for law enforcement, the TLP model increases the chances of cultural acceptance while also ensuring compliance with regulations such as the EU's AI Act and the UK's Covenant for Using Artificial Intelligence in Policing.
Why policing needs its own AI framework
Until now, most AI governance models have focused on general ethics or technical standards. These are valuable but often too broad to address the unique challenges of law enforcement. Policing decisions directly affect people's freedoms, rights, and even life-and-death situations. A flawed risk assessment or biased investigative tool could undermine public trust and lead to serious legal consequences.
Halford argues that the integration of generative AI cannot simply mirror how it is adopted in healthcare, finance, or education. In policing, public legitimacy is just as important as technical performance. The TLP model therefore emphasizes stakeholder engagement, legal scrutiny, and ethical review alongside technical testing. It is designed as much to build confidence as to improve efficiency.
The promise and the limits of generative AI
The article outlines the broad spectrum of tasks where generative AI might be applied. In theory, it could support officers by answering emergency calls, drafting witness statements, analyzing vast sets of intelligence, planning interviews, or even generating training materials. At the administrative level, it could summarize documents, manage schedules, or automate routine reporting, freeing officers for frontline duties.
Yet these possibilities come with risk. When generative AI was tested on UK police risk assessments, it produced results that were only about 62 percent accurate - far too low for operational deployment. The technology has also shown a tendency to "hallucinate," producing outputs that look convincing but are factually wrong. These issues underline why the TLP model insists on rigorous testing, clear performance thresholds, and ongoing human oversight before any system is put into use.
Balancing potential and public trust
The deeper challenge is not only technical but cultural and ethical. Generative AI can analyze data and automate processes at a scale no human team could match, but without careful safeguards it risks amplifying existing biases or eroding trust between police and the communities they serve. For this reason, the TLP framework builds legitimacy into its core. It requires public consultation, fairness testing, and open disclosure of how the technology is being used.
If policing is to benefit from the efficiencies of AI, it cannot ignore the lessons of past controversies over surveillance and predictive algorithms. The future of AI in law enforcement depends on ensuring that every innovation complements officers rather than replaces them, and that transparency and accountability are never sacrificed for speed.
A turning point for AI in policing
The introduction of the Transformer Led Policing model marks a turning point in the debate. The question is no longer whether AI can be used in policing, but how it should be used and under what rules. If adopted carefully, AI could reduce paperwork, sharpen investigations, and improve safeguarding. If deployed recklessly, it risks repeating old mistakes and damaging public confidence.
Halford's framework offers a way forward: a methodical, police-specific roadmap that balances technological opportunity with ethical obligation. In the end, the success of AI in policing may not hinge on what the technology can do, but on whether the public believes it is being used fairly, responsibly, and in the service of justice.
