The NYPD Matched the Wrong Face and Jailed an Innocent Man for Two Days

What happened

In April 2025, Trevis Williams, a New York City man, was arrested and held in jail for more than two days for an indecent exposure incident he had nothing to do with. The NYPD's facial recognition system had matched his face to surveillance footage from a Manhattan case. The match was wrong. Williams had physical characteristics that did not fit the suspect description, and he had a verifiable alibi. Neither fact stopped his arrest.

The NYPD ran its facial recognition tool against low-quality, grainy CCTV footage from the scene. The algorithm returned a match to Williams' mugshot from a prior, unrelated arrest, giving investigators a name to pursue. The facial recognition reports themselves noted the match did not constitute probable cause. That warning did not change what happened next. Investigators moved toward an arrest anyway, and a victim's misidentification added enough apparent confirmation to proceed.

The alibi Williams offered was checkable. His phone location data and employer records could have placed him elsewhere at the time of the incident. Neither was verified before the arrest. He was jailed for over two days before the failure became apparent. Charges were eventually dropped several months later, but by then Williams had already experienced loss of liberty, reputational harm, and the serious threat of being placed on a sex offender registry.

The technology's limitations in situations like this are not new or unknown. Facial recognition algorithms trained predominantly on certain demographics carry documented accuracy gaps when applied to faces outside those groups, and low-resolution source images push error rates higher still. This case sits within a documented pattern: the NYPD's use of the same technology has drawn scrutiny in prior wrongful arrest cases involving men of colour, including Robert Williams and NiJeer Parks, and advocacy organizations have called for an outright ban on its deployment in law enforcement until the accuracy and accountability gaps are resolved.

What makes the Williams case structurally distinct from a simple misidentification is the sequence of ignored checkpoints. A report flagged the match as insufficient for probable cause. Physical characteristics did not align with the suspect. An alibi was offered and not checked. Each was a point where a verification step could have changed the outcome, and none of them did. The accountability gap here is not only about the algorithm: it is about the absence of a provable record of what the system returned, how investigators weighted it against the warnings, and who approved moving forward despite them. Without that record, each wrongful arrest gets treated as an isolated failure rather than evidence of a systematic one.

Reported impact

Affected parties

Not publicly disclosed

Harm type

Not publicly disclosed

Scale

Not publicly disclosed

Financial impact

Not publicly disclosed

Regulatory action

Not publicly disclosed

Classification

Organization

Not publicly disclosed

AI system

Not publicly disclosed

Industry

Not publicly disclosed

Country

Not publicly disclosed

Provider

Not publicly disclosed

Incident type

Not publicly disclosed

Relevant governance controls

Governance control mapping is not available for this record.

• No controls mappedNot publicly disclosed

Control mapping is analytical. It does not state that any control would have prevented the incident.

Sources and evidence

This record was researched and written by the Index. The event is also catalogued in the following database, which is listed for cross-reference.