INSIDE THIS REPORT

• The FBI and West Virginia University have studied twins at an Ohio festival for over 15 years
• Identical twins expose the hardest edge case in facial recognition technology
• Strict privacy safeguards keep the research fully separated from FBI investigative databases

By Samuel López | USA Herald

Every summer, thousands of identical strangers converge on a small Ohio town — and federal researchers are waiting for them.

At the Twins Days Festival in Twinsburg, Ohio, alongside the funnel cakes and matching-outfit contests, sits a research corridor regulars call "Research Row." For more than fifteen years, the FBI and West Virginia University have set up there, recruiting twin volunteers for a study built around one of the toughest unsolved problems in modern security technology: how do you tell apart two people who are, biologically, near-mirror images of each other?

The stakes are bigger than a science fair curiosity. Facial recognition, iris scanning, and fingerprint matching now sit behind everything from unlocking a phone to clearing airport security to identifying a suspect in a criminal investigation. And identical twins represent the single hardest test those systems can face.

"From a biometrics perspective, identical twins are the ultimate challenge because we are trying to uniquely identify one person from another," said Ben Smith, a biometric program manager in the FBI's Criminal Justice Information Services Division. Facial recognition in particular struggles here — twins can look, to an algorithm, essentially indistinguishable. Fingerprints and iris patterns, by contrast, remain unique even between identical siblings, giving researchers a reliable baseline to measure against.

That gap between "looks the same" and "is actually unique" is where the real work happens. By running facial recognition, fingerprint, and iris technologies against twin data, researchers can stress-test algorithms under conditions ordinary datasets simply can't replicate. If a system can correctly separate identical twins, its accuracy across the general population becomes far more trustworthy.

What makes the dataset especially valuable isn't just its difficulty — it's its depth. Many twin families return to the festival year after year, giving researchers something exceedingly rare in biometric science: longitudinal data tracking the same individuals over more than a decade.

"Very little longitudinal data exists today," Smith said. Because the study has run for so many years, researchers aren't just testing whether current algorithms can distinguish twins — they're watching how faces, and the AI systems reading them, change over time.

That long view has taken on new weight as artificial intelligence reshapes the field entirely. "Now, in 2026, data is everything," said Jeremy Dawson, a WVU computer science and engineering professor leading the study. Training a neural network properly requires millions of images — and a population of people who already look nearly identical gives researchers an efficient shortcut to the hardest possible test case.

The applications reach well past the festival tents. Biometric verification now secures financial transactions, unlocks personal devices, and confirms identity at borders. In law enforcement, it helps investigators identify unknown individuals and corroborate forensic evidence. Smith noted that when facial recognition is the only lead examiners have, they must factor in the possibility of an identical twin skewing results — a limitation this research helps quantify and correct for.

None of it happens without guardrails. The project operates under the FBI's Institutional Review Board and full compliance with the Common Rule, the federal ethical framework governing human-subjects research. Every participant volunteers and gives informed consent. West Virginia University strips identifying information and assigns anonymized ID numbers before any data reaches the Bureau — and per senior FBI policy advisor Michael Yates, that information is walled off entirely from investigative or operational use.

"The FBI IRB generally prohibits any use of the information provided by research participants in FBI investigations or operational databases," Yates said.

The result is a rare thing in modern AI development: a system being pressure-tested against its hardest possible failure case, with the humans behind the data fully protected in the process. As Smith put it, the goal isn't just proving what the technology can do — it's mapping precisely where it still falls short, so the tools are never trusted beyond their actual limits.

ABOUT THE AUTHOR

Samuel López is a Senior Legal Analyst and investigative journalist whose two-decade career has focused on the legal frameworks governing government data collection, forensic evidence standards, and emerging technology oversight. His background examining how federal agencies balance investigative tools against privacy and civil liberties protections gives him a grounded, credentialed lens for reporting on the intersection of AI, biometrics, and law enforcement policy.

Stay Ahead of the Story

If a story like this one caught your attention, our newsletter is built for readers like you. Get investigative reporting before it breaks, sharp analysis on the legal and technological forces shaping national security, and access you won't find anywhere else. Subscribe to the USA Herald newsletter and stay ahead of it.

👉 Follow us on X: https://x.com/RealUSAHerald