Source: Financial Times (paywall)
Britain's National Police Chiefs' Council has explicitly prohibited forces from deploying generative AI to draft evidence summaries and court statements, abandoning an efficiency fix that risked hallucinations contaminating criminal prosecutions. This is one of the first major institutional retreats from AI deployment in high-stakes legal settings—not due to regulatory gaps but because the operational cost of AI errors (wrongful convictions, collapsed cases) outweighs time savings. Sectors with liability exposure and procedural rigor are hitting a practical ceiling on large language models faster than hype cycles predicted.