Use of Facial Recognition in Court

Facial recognition technology (FRT) has moved from consumer authentication into criminal investigation faster than courts have developed a consistent framework for evaluating it. This paper synthesizes four legal and policy case studies-Dillon v. City of Jacksonville Beach, Williams v.

City of Detroit, Patel v. Facebook, and State v. Miles-to ask a narrower question than whether facial recognition is simply 'accurate' or 'inaccurate': under what conditions can facial recognition be used in a justice system without undermining reliability, due process, privacy, and the ability of defendants to challenge the basis of state action?

Across the four cases, the recurring problem is not a single defective algorithm. It is a chain of interacting risks: poor probe images, demographic differentials, similarity scores misunderstood as identification probabilities, automation bias, suggestive follow-up procedures, incomplete disclosure, proprietary systems, and broad biometric databases. The case comparison shows that errors become most dangerous when an FRT lead is allowed to contaminate later investigative steps and then disappear from view before trial.

Williams demonstrates the human consequences of a false match and the value of mandatory corroboration; Dillon shows how an apparently precise score can anchor investigators despite contradictory facts; Miles establishes the importance of discovery even when prosecutors do not intend to introduce the FRT result itself; and Patel supplies a privacy framework for treating face geometry as a legally significant biometric interest rather than ordinary public information. The synthesis supports a conditional-use model rather than either unrestricted adoption or a purely technical ban: FRT should remain an investigative lead, never a standalone basis for arrest or conviction; material details of the search should be disclosed; independent evidence should precede lineups and warrants; systems should be independently tested under real-world conditions; and courts should treat secrecy claims skeptically when they prevent meaningful challenge. The central conclusion is that the legitimacy of facial recognition in court depends less on whether an algorithm can generate a candidate than on whether the legal process preserves contestability after it does.