Facial Recognition Used in Suspect Identification

Facial recognition is increasingly used by U.S. law enforcement agencies to generate investigative leads and identify criminal suspects. While proponents emphasize efficiency and public safety benefits, federal reviews and empirical testing show that these systems exhibit uneven performance across demographic groups and operational conditions, and that weak procedural controls can allow probabilistic matches to function as de facto evidence. A growing number of wrongful arrests linked to facial recognition illustrate a recurring governance failure: low-quality or misinterpreted matches are treated as determinative without adequate corroboration, disclosure, or accountability. Despite these risks, the United States lacks a uniform federal baseline governing how facial recognition may be used for suspect identification, how such use must be documented and disclosed, and how agencies should be audited for compliance and performance. This paper proposes a mechanism-grounded federal framework for regulating facial recognition used in suspect identification, built around three pillars: disclosure, accountability, and independent auditing. It recommends conditioning key Department of Justice law-enforcement grants on agency compliance with standardized requirements for case-level documentation, court disclosure, corroboration of facial recognition leads, supervisory sign-off, and operator training. It further proposes a national audit protocol aligned with NIST testing concepts and Government Accountability Office findings, coupled with a public reporting database maintained by the Department of Justice. By grounding regulatory authority in existing grant mechanisms and standardized reporting and audit procedures, the framework establishes a realistic and enforceable federal floor of safeguards. The proposed approach preserves facial recognition as an investigative lead tool while reducing the risk of wrongful arrest, civil-rights violations, and unaccountable reliance on probabilistic identification technologies.