AI in Policing: Risks and Benefits

0:00It's April 6th, this is Vancouver News, today local news powered by AI. The Royal Canadian Mounted Police kicked off a year-long pilot back in July, testing AI software called Draft1 from AXON in a British Columbia. The attachments and two in Alberta. They budgeted up to $200,000 for it, using body camera audio to auto-generate draft incident reports and cut down on paperwork. Time. This is the first public nod to the project in their recent 2026-2027 departmental plan. Officers upload just the audio no video or facial recognition by some other forces, and they have to edit at least 10% of every draft. Fix. Obvious errors and electronically sign off on its accuracy before submitting. Experts are raising red flags, saying AI lacks the human discretion officers bring from what they see and analyze on the scene. This worry about hallucinations, where the text spits out fake details that could end up in court and question of police rushing into this.

1:02Without weighing real-world risks to people's lives, AXON claims the tool, built on an open AI model, gives officers a head start by saving about one hour per shift with built-in prompts to add details, and tweaks to curb wild inventions. But a Utah police force recently caught a hallucinating and officer turning into a frog from a movie playing nearby. The pile is still rolling, with evaluations underway to see if it truly frees up cops for street work over desk duty, but watch dogs like that. Electronic Frontier Foundation argue it dodges accountability since it's hard to spot AI's fingerprints in final reports.