IntelliVision facial recognition: the FTC order on accuracy and bias claims

What the public record shows

The FTC finalized an IntelliVision consent order in January 2025, settling allegations about unsupported accuracy, lack-of-bias and anti-spoofing claims. The order prohibits misrepresentations and requires competent and reliable testing for the specified effectiveness claims.

Where the promise broke

A single accuracy score can conceal differences between recognizing a known person and deciding that an unknown person matches a database entry. Our procurement checklist separates those tasks and requires the intended threshold. Ask what population the evaluation covers, which input conditions were tested and how incorrect matches are handled. Ideal-lighting demonstrations are weak substitutes for documented tests reflecting the actual customer journey.

What this record does not establish

This is a civil regulatory consent order, not a criminal conviction. The announcement describes allegations settled by the order; it does not establish performance for every deployment or certify competing products. Accuracy, demographic performance and resistance to spoofing are separate questions and should not be condensed into a generic secure-AI badge.

The buyer lesson

Decide whether facial recognition is necessary for the task. Request evaluation conditions, an alternative access method and a correction process. Test failure cases with authorized participants and approved data. Record the consequence of a false match, not just successful matches. Keep procurement, privacy and security review connected to the implementation rather than letting marketing decide the access policy.

Sources: FTC final-order announcement · January 2025