A Facial Recognition Error Put an Innocent Man in Jail for Two Days

What happened

Facial recognition technology has been promoted to law enforcement agencies as a tool that narrows investigations, not one that ends them. The arrest of Trevis Williams in connection with a Union Square indecent exposure case shows what happens when that distinction collapses.

The NYPD's facial recognition system returned Williams as a match for the suspect. Officers arrested him, booked him, and held him for more than two days. He was formally charged. What the system had not communicated, and what the investigation apparently did not resolve before the arrest, was the information that should have disqualified him. There were reportedly notable physical differences between Williams and the person described in the original complaint. Williams also had phone data placing him elsewhere at the relevant time. The charges were eventually dismissed.

The word "allegedly" recurs throughout accounts of this case because no proceeding has fully established where the failure originated or who made which call. But the structural problem does not depend on resolving individual fault. A facial recognition match is a probabilistic output. It suggests a candidate. It does not establish identity, and it does not override physical evidence pointing in the other direction. When an arrest proceeds on that output anyway, something in the review process either failed to catch the discrepancy before it mattered or was not designed to.

Williams is not the first person jailed after a facial recognition system returned an incorrect match. Documented wrongful arrests tied to the technology have accumulated across multiple jurisdictions over the past several years. In each case, the common thread is not a system that failed to produce a result but a process that treated the result as more definitive than it was. Review steps that should have introduced doubt before an arrest were either absent, cursory, or simply did not carry weight.

That is the gap this case makes visible. There is no standardized requirement for law enforcement agencies to document how a facial recognition output was weighed against other evidence, who reviewed the comparison, and what conditions had to be met before an arrest decision was made. A provable record of what a system did, and what human checks were completed before a person was taken into custody, would allow investigators and oversight bodies to trace exactly where the process failed. Without that record, the same error runs again, and the correction arrives too late to give back the days already lost.

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.