FBI Uses Twins to Test Limits of AI Biometric Systems

The FBI has detailed a long-running research program that uses biometric data from twins to test how reliably artificial intelligence systems can distinguish people with unusually similar appearances.

The bureau has worked with West Virginia University for more than 15 years at the annual Twins Days festival in Twinsburg, Ohio. Hundreds of volunteers have contributed facial photographs, fingerprints and iris images, creating a dataset that lets researchers study both close physical resemblance and changes in biometric characteristics over time.

Identical twins provide a demanding test for face recognition because they share many visible features. The research has found that fingerprints and irises remain individually distinctive, giving system designers additional signals when facial similarity makes a single-modality decision more difficult.

The longitudinal component is also important. Participants who return in later years allow the research team to measure how aging, appearance changes and capture conditions affect matching performance. The resulting data can be used to evaluate new algorithms and identify failure patterns before systems are considered for operational use.

Combining several signals is a common approach in high-assurance identification. NIST has released multimodal research datasets containing linked face and fingerprint records, while independent benchmarks continue to measure fingerprint and iris algorithm performance under controlled conditions.

The FBI says the Twins Days project operates separately from law-enforcement investigations. Participation is voluntary and subject to informed consent, the study is reviewed by the bureau’s institutional review board, and each participant’s records are assigned a random identifier rather than being stored under a name.

The research data are not added to operational biometric databases or used to identify people in investigations. Access is limited to approved researchers working on the project, and results are evaluated as scientific evidence rather than treated as an automated identity decision.

The bureau also emphasizes that biometric algorithms support human examiners rather than replace them. A system can narrow candidates or flag similarities, but trained specialists review the underlying images and other evidence before reaching a conclusion. Studying twins helps establish where automated matching is strongest and where corroborating modalities and expert review are most valuable in practice.

Sources: Federal Bureau of Investigation

–

By Ali Nassar-Smith