A New York court found NYPD misused facial-recognition AI, leading to false imprisonment

A New York Criminal Court found in People v Zuhdi A. that NYPD and FDNY officials used unauthorized facial recognition software (Clearview AI) instead of the approved limited database, illegally accessed DMV records without a court order, and altered a defendant photograph by modifying neck length before placing it in a photo array. The same pattern of misuse caused Trevis Williams to be falsely arrested and jailed for two days despite not matching the physical description and being miles away at the time of the crime. Both cases were ultimately dismissed.

New York Police Department (NYPD) · Incident Aug 1, 2025 · Indexed Jun 4, 2026 · 3 sources

Clearview AI returned hundreds of matches from an uncontrolled web-wide search, and investigators altered a suspect DMV photo before placing it in a lineup with no documented justification for either step.
What
A New York Criminal Court found in People v Zuhdi A.
Incident date
Aug 1, 2025
Who
New York Police Department (NYPD)
Failure mode
Tool Misuse
AI surface
Search / RAG
Severity
High

What happened

In People v Zuhdi A., decided June 17, 2025, a New York Criminal Court found that law enforcement extracted still images from video surveillance, used unauthorized facial recognition software (Clearview AI) instead of the NYPD approved limited database, illegally accessed DMV records to obtain a driver license photo without a court order, and altered the defendant photograph by changing the length of his neck before placing it in a photo array for witness identification. The same pattern of misuse led to the false arrest and two day jailing of Trevis Williams, who did not match the physical description of the suspect and was miles away at the time of the crime. Prosecutors dismissed the Williams case after the Legal Aid Society proved he was falsely identified, and the Zuhdi A. case was dismissed on speedy trial grounds after the court found 187 days of prosecutorial delay tied to failure to disclose the misused facial recognition records.

What broke inside the model

Failure path · mode profile · Tool Misuse
  1. 01 · TriggerThe agent selects the correct tool.
  2. 02 · Model stepIt fills the call with the wrong arguments.
  3. 03 · Control gapNo validation checks the arguments first.
  4. 04 · FailureThe tool runs against the wrong target.
  5. 05 · ConsequenceThe wrong record, account, or system is hit.

At the tool call, the arguments point at the wrong target.

Clearview AI returned hundreds of potential matches from an uncontrolled web-wide search with no documented or objective method for how investigators narrowed results to a single suspect. NYPD officers circumvented their own policies by routing searches through the FDNY and using unauthorized databases, while physically altering suspect photographs before presenting them in identification arrays. The absence of transparency and documentation at every step meant a false match could not be caught or audited before it led to an arrest.

Public visibilityHigh
Regulatory exposurePossible
Customer impactClass-wide
Financial impactUnknown
Time to disclosureMonths
  1. Court FilingPeople v Zuhdi A. (2025 NY Slip Op 51047(U))nycourts.gov
  2. PressMan's wrongful arrest puts NYPD's use of facial recognition surveillance tech under scrutinyabc7ny.com
  3. PressHow the N.Y.P.D.'s Facial Recognition Tool Landed the Wrong Man in Jailnytimes.com
Permalinkhttps://failureindex.ai/failures/new-york-court-found-nypd-misused
CitationAI Failure Index. "A New York court found NYPD misused facial-recognition AI, leading to false imprisonment" (FI-0112). Realm Labs. https://failureindex.ai/failures/new-york-court-found-nypd-misused (indexed Jun 4, 2026).
Share cardA branded image of this record for posts and slides.

Data fields CC-BY 4.0, prose citation permitted. Incident ID FI-0112. Full dataset at /data.

Note from Realm Labs, the Index steward

How Realm would have caught this

Controls for this failure mode
  • OmniGuard
  • AgentRealm

Realm can inspect a tool call against the user's actual intent before it runs, and hold calls whose arguments or target do not match what was asked, so the wrong tool or the wrong arguments never reach the system of record.