Researchers: NYPD used edited suspects' photos and photos of celebs, when witnesses said the suspects looked like them, to generate facial recognition matches
In one case, a photo of the actor Woody Harrelson was used to locate a suspect — A new report from Georgetown Law's Center …
Context & Ripple Effects
Georgetown Law's Center on Privacy & Technology documents that the NYPD did not just run real probe photos through facial recognition — when witnesses said a suspect resembled someone else, detectives fed in edited suspect photos or celebrity stand-ins like Woody Harrelson, meaning the system was matching on manipulated inputs rather than observed faces. That practice sits alongside what records later showed about the department's database hygiene: it had been uploading arrest photos of juveniles since 2015 despite higher false-match risk in young faces.
The report lands early in an arc that has since hardened into a pattern: NYPD's own disputed claim about identifying a terrorism suspect via Clearview AI, documented wrongful arrests driven by false positives and automation bias, and a 2025 investigation finding 15 departments across 12 states arresting suspects on facial recognition alone. What looked like an isolated technique in 2019 now reads as one node in a national accountability problem.
First-order effects
- Suspects identified through these doctored or celebrity-substitute probes face arrests built on matches their own faces never produced — the Georgetown report makes the NYPD's input practices a matter of public record and litigation-ready discovery material.
Second-order effects
- Every department relying on vendor facial recognition now inherits the credibility problem: if the NYPD matched on Woody Harrelson, prosecutors and defense counsel elsewhere can challenge any match generated from non-standard probe images, forcing departments to document their inputs.
Third-order effects
- If the pattern holds — manipulated inputs feeding matches, matches treated as sufficient evidence for arrest — the pressure point shifts from the algorithms to police procedure itself, pushing toward rules requiring corroborating evidence beyond a facial match and warrant-style oversight of probe image creation.
The trend: Facial recognition policing is moving from a tool-of-last-resort narrative to a documented systemic practice, with each investigative report narrowing the space in which departments can claim the technology is used responsibly.