Law Report How accurate is facial recognition software?
Mar 31, 2026
Associate Professor Catherine Kemp, a law and tech policy leader, outlines legal and privacy concerns with facial recognition. Alvi Chowdhury, a software engineer wrongly arrested after a match, shares his personal account. They discuss wrongful arrests, racial bias and error rates, retailers' use of real-time systems, failures in notification and oversight, and calls for stronger limits and human review.
AI Snips
Chapters
Transcript
Episode notes
Wrong Arrest Caused By A Flawed Match
- Alvi Chowdhury was wrongly arrested after a UK police facial recognition match linked his four‑year‑old custody photo to CCTV from a burglary in Milton Keynes.
- He had alibi evidence, was 150 km away, and only discovered his custody photo remained on police systems despite an earlier NFA.
Facial Recognition Has Higher Error Rates For Minorities
- The UK Home Office research shows markedly higher error rates for ethnic minorities: about 4% for South Asian males and 9% for Black females.
- Alvi says police still deploy systems with error rates tech companies would call catastrophic.
Retailer Used Real‑Time Scanning Against Offender List
- Bunnings used real‑time facial recognition across about 60 stores to screen customers against a list of repeat offenders and claimed nonmatches were deleted almost immediately.
- If matched, staff reviewed captured images to decide whether to remove the person.