Question
Can the efficiency of PoliceAI justify its rollout ahead of a dedicated legal framework, adequate safeguards for confidentiality, and the independent development of accurate, unbiased domestic AI systems?
Short Answer
While AI can ease police administration, such as summarising case files and easing funding pressures, PoliceAI is not governed by dedicated legislation. It has also failed to fulfil its original purpose of centralising AI use, leaving force-by-force guidance too vague to protect sensitive information or guarantee unbiased analysis from non-domestic AI systems.
Facts
Since 2020, Artificial Intelligence has been used increasingly across government institutions. In 2023, the National Police Chiefs’ Council (NPCC) reported that all UK police forces were using AI, but without official training or secure frameworks.
The AI Opportunities Action Plan (2025) and the Police Reform White Paper, published under the Starmer government, attempted to centralise AI use across all 43 police forces through PoliceAI, launched in June as a national centre for responsible AI in policing.
Guidance included a Responsible AI Checklist and ethical provisions. However, since Andy Burnham became Prime Minister, the programme’s priority has fallen. The Department for Science, Innovation and Technology was dissolved, and the UK’s first AI minister, Kanishka Narayan, was appointed without a dedicated department, leaving PoliceAI at risk of losing prominence just as it needs stronger oversight.
Analysis
A Regulatory Vacuum
PoliceAI’s efficiency gains are notable: 800 hours of footage in a kidnapping case were reviewed in three hours, producing an early guilty plea, while AI transcription of 999 and 101 calls has helped link crime reports.
However, these gains rest on shaky foundations. Section 57 of the Police Act 1996 and the NPCC’s checklist fall short of binding legislation, leaving no uniform, enforceable standard. Every AI tool risks hallucinations, opaque “black box” decision-making, and the disclosure of sensitive data. Existing guidance flags these risks without offering practical safeguards or a remedy scheme for those affected.
The Cost of Outsourced AI
Despite obligations under the Equality Act 2010 and Data Protection Act 2018, bias persists. Systems rarely explain their reasoning, and demographics underrepresented in training data are reflected in skewed outputs.
Most AI used by UK police, including Microsoft Copilot, is foreign-built, raising accuracy and cybersecurity concerns. Brazil’s APOIA, by contrast, was built domestically to reflect its own courts. Without equivalent UK legislation, forces are exposed to risks they have neither anticipated nor built the capacity to manage.
Facial Recognition and Predictive Policing
Facial recognition, used to confirm identity, is more straightforward than predictive policing, which forecasts behaviour and is far more legally contentious. The Metropolitan Police recorded 10 false alerts from predictive policing out of 2,077 uses, four confirmed as false positives only after arrest. Its need for continuous monitoring may also breach Article 8 of the European Convention on Human Rights, which protects the right to privacy.
Practical Implications
Specialist practices are emerging to challenge biometric and AI-generated evidence. Lawyers will help shape future legislation, defining lawful boundaries under human rights law and advising on data protection and equality impact assessments.
Future Outlook
Without a dedicated statutory framework, PoliceAI’s expansion will likely face mounting legal challenges over bias and privacy. Much depends on whether the Burnham government revives centralised oversight or allows individual forces to keep developing AI use in an ad hoc, legally precarious way.