A Derbyshire Police officer is under criminal investigation for allegedly using artificial intelligence to create false evidence in multiple criminal cases, Sky News reported on 13 June 2026. The Crown Prosecution Service is advising Derbyshire Constabulary on the inquiry. No arrests have been made.

The officer has been removed from active duty while investigators proceed. If the allegation is proved, it would constitute perverting the course of justice under English law — a charge that carries a maximum life sentence, underscoring how seriously courts treat any tampering with the integrity of criminal proceedings.

The early involvement of the Crown Prosecution Service is significant. Prosecutors typically join investigations in an advisory role before charges are laid, helping police develop lines of inquiry that will survive scrutiny at trial. Their presence here suggests investigators are already examining whether specific cases, verdicts, or convictions were materially compromised by the alleged conduct.

The exact AI tools used have not been publicly disclosed — and that absence matters. Different types of generative AI pose different forensic problems. Text models can produce plausible witness statements or transcripts; image and video models can synthesise photographs or CCTV-style footage; audio models can replicate voices. Each category creates a distinct challenge for investigators trying to separate authentic evidence from fabrication within case files.

Detecting AI-generated evidence after the fact is not yet a routine process. Forensic watermarking standards — such as those being developed under the C2PA (Coalition for Content Provenance and Authenticity) framework — are designed to work prospectively on newly created content, not to identify older fabricated material. Statistical detectors that look for tell-tale signs of AI generation have meaningful error rates, and they become less reliable as the underlying AI models improve. Any review of cases this officer touched will require careful, case-by-case examination rather than a single automated check.

UK policing has faced sustained scrutiny around evidence standards and disclosure in recent years. The Post Office Horizon scandal — though in a different sector — sharpened judicial and public attention to what happens when institutions tolerate systemic failures. An individual officer acting deliberately is a different category of problem than institutional failure, but the effect on convicted defendants is comparable: their convictions rest on manufactured evidence.

This incident falls into territory that regulatory bodies have flagged in theory for some years. The College of Policing has been developing guidance on AI use in law enforcement, and the National Police Chiefs' Council has worked to define acceptable and unacceptable applications. Those frameworks were built primarily around algorithmic decision-support tools — predictive analytics, facial recognition — not deliberate misuse of generative AI by individual officers. This case exposes a gap in that framework.

The investigation covers multiple cases, meaning consequences could extend well beyond any sanction against the officer himself. Defence solicitors representing clients convicted in cases this officer worked on will have clear grounds to seek case review through the Criminal Cases Review Commission. The CCRC already operates under substantial caseload pressure; a wave of referrals tied to one officer's alleged conduct could strain a body with limited capacity.

Derbyshire Constabulary has not publicly disclosed how the alleged conduct was discovered, over how long a period it occurred, or how many cases are under review. Those details will be crucial — both for any prosecution and for understanding the scale of the remediation effort ahead.

The foundation of criminal justice is the presumption that evidence presented in court is genuine. Generative AI's capacity to produce convincing synthetic material cheaply and with minimal technical skill alters the threat environment for that presumption — not just from external bad actors but, as this case suggests, from within institutions.