Summary

• A Derbyshire Police officer is being investigated over allegations involving AI created evidential material in multiple cases.
• The Crown Prosecution Service has said it is working with Derbyshire Police and engaging with affected defence teams and courts.
• The case sharpens the governance challenge around AI in policing, especially where tools touch evidence, disclosure, statements, or court processes.

The Crown Prosecution Service is working with Derbyshire Police after allegations that an officer used artificial intelligence to create evidential material in a number of cases, raising a sharper governance problem for AI in policing than ordinary productivity claims.

The officer is being investigated over accusations involving AI created evidence, while the CPS has said it is engaging with defence teams and courts that may have been affected. Derbyshire Police has said the officer was removed from frontline duties while inquiries continue.

The allegations remain under investigation, and no conclusion should be drawn about the officer’s conduct before the process is complete. Even so, the case exposes a high consequence boundary for generative AI in public services. Drafting administrative material is one thing. Creating or altering evidential material, witness related documents, disclosure schedules, or court paperwork enters a different risk category.

The Home Office has published a police AI factsheet describing the new National Centre for AI in Policing, known as PoliceAI, which began mobilising in April and was formally launched in June. That centre is intended to consolidate work on AI adoption across policing. The Derbyshire case shows how quickly the agenda can move from efficiency to evidential integrity.

Auditability is the hard control

Generative AI systems can produce plausible language without preserving a reliable chain of reasoning or evidence. In criminal justice, that weakness is not a technical inconvenience. Courts depend on provenance, disclosure, accuracy, and accountability. A document used in a case must be traceable to its source material and to the person responsible for producing it.

AI use in policing may still have legitimate value. Tools can help process large volumes of CCTV, organise digital evidence, transcribe material, redact sensitive information, summarise non-evidential documents, or reduce the time officers spend on repetitive administrative tasks. Those uses require clear boundaries, version history, disclosure rules, and systems that separate generated assistance from evidential fact.

The risk is not only that an AI system may hallucinate. A human user may trust, edit, or present generated output without enough scrutiny, while later reviewers may be unable to distinguish source evidence from AI created text. In a criminal case, that can affect defendants, victims, witnesses, prosecutors, police credibility, and the courts.

Public sector AI programmes often begin with a productivity promise: fewer hours spent on paperwork, faster case preparation, and more time for frontline work. Those benefits are attractive in policing, where workloads are heavy and digital evidence volumes continue to rise. The Derbyshire allegations show why governance cannot be deferred until after deployment.

AI policy meets operational practice

The national policing AI agenda will need to answer practical questions. Which tasks can officers use AI for? Which tools are approved? What must be disclosed? How are outputs logged? Who checks accuracy? What happens if a document has been AI assisted? How are prompts, drafts, edits, and final versions retained?

Those questions are not solved by broad statements about responsible AI. They require standard operating procedures, training, technical controls, procurement rules, and audit systems that work under daily operational pressure. If officers reach for public or commercial AI tools because official systems are slow, unavailable, or unclear, policy has already failed at the point of use.

The case also has relevance beyond policing. Public bodies across government are adopting generative AI to reduce administrative pressure, draft documents, triage casework, and summarise records. The more consequential the process, the more the organisation needs evidence of how AI was used and who remains accountable for the final output.

Policing sits at the hardest end of that spectrum because errors can affect liberty, safety, and public trust. AI can still support criminal justice work, but only where the system preserves provenance and human accountability. The Derbyshire investigation will be watched because it tests whether public sector AI governance can cope when the technology enters the evidential chain rather than the back office.