Yes, federal law enforcement agencies have been sued for using facial recognition algorithms on Black defendants without adequate validation, leading to multiple wrongful arrests. The FBI can search against its massive database of at least 640 million images of U.S. adults using facial recognition technology, but the algorithms powering these searches have been repeatedly shown to have significantly higher error rates for Black people compared to white people.
Robert Williams of Detroit became one of the first documented victims when he was arrested in 2020 based on a facial recognition match that turned out to be incorrect—a case that settled for $300,000 in June 2024. Since then, numerous other Black men and women have filed lawsuits after being wrongly arrested based on unreliable facial recognition identifications, prompting civil rights organizations to challenge the FBI and other federal agencies’ use of this technology.
Table of Contents
- Why Are Facial Recognition Algorithms More Likely to Misidentify Black People?
- The Pattern of Wrongful Arrests Since 2018
- The FBI’s Massive Facial Recognition Database
- Settlement Amounts and What They Reveal
- The Critical Lack of Due Process Protections
- The ACLU’s Federal Lawsuit
- The Road Ahead and Calls for Algorithm Validation
Why Are Facial Recognition Algorithms More Likely to Misidentify Black People?
The racial bias in facial recognition is not accidental—it’s a direct result of how these algorithms were trained. Most facial recognition algorithms used in the United States were built using datasets consisting primarily of white faces, leaving them poorly equipped to accurately identify people of other races. According to a National Institute of Standards and Technology (NIST) study, Black and Asian people are 10 to 100 times more likely to be misidentified by facial recognition than white people. This massive discrepancy in error rates means that when police and federal agencies run facial recognition searches, they are far more likely to get false matches when searching for suspects who are Black or Asian.
The problem extends beyond just raw error rates. Higher false positive rates have been documented for African Americans, Asians, and Native Americans compared to Caucasians across U.S.-developed facial recognition systems. This is not a minor technical flaw—it’s a systemic vulnerability that can destroy lives. When a facial recognition algorithm with a 10 to 100 times higher error rate for Black individuals is used to justify an arrest, the consequences are immediately apparent: wrongful detention, trauma, legal costs, and lost employment.
The Pattern of Wrongful Arrests Since 2018
Since 2018, at least six Black men and one Black woman have been subjected to days in jail after being misidentified by facial recognition technology. These are not isolated incidents; they represent a documented pattern across multiple states. law enforcement departments in Louisiana, Maryland, Michigan, and New Jersey have all been accused of making false arrests based on facial recognition matches, suggesting a widespread reliance on this technology without adequate safeguards. The most prominent case involved Robert Williams, who was arrested in Detroit in 2020 after facial recognition software incorrectly matched his photograph to a theft suspect. Williams spent days in custody before the error was discovered.
His wrongful arrest became a turning point in public awareness of facial recognition bias. However, what’s crucial to understand is that Williams’ case is not unique—it’s part of a larger pattern that continues today. Another documented case involved Porcha Woodruff, who was falsely arrested while eight months pregnant based on an unreliable facial recognition match. These cases demonstrate that the problem affects both men and women and can occur regardless of the defendant’s circumstances. More recently, Jason Killinger’s case is heading toward a 2026 trial after he was wrongly detained for 11 hours due to a facial recognition error, showing that wrongful identifications continue to occur years after the initial warnings about algorithm bias.
The FBI’s Massive Facial Recognition Database
The FBI’s facial recognition database is not limited to mugshots of convicted criminals or even people arrested for crimes. The FBI can match or request matches against at least 640 million images of U.S. adults. This staggering database includes driver’s license photos, passport images, and other governmental identification documents, meaning that virtually any American could potentially be matched by the FBI’s facial recognition system, even if they have never been arrested or charged with a crime.
The sheer size of this database creates a critical problem: when searching such an enormous collection using an algorithm with inherent racial bias, the number of false matches increases dramatically. A biased algorithm searching through hundreds of millions of faces is far more dangerous than the same algorithm searching through a smaller, more targeted database. The FBI has not publicly disclosed how many searches it conducts, how often false matches occur, or what protocols it follows to verify facial recognition matches before using them in criminal investigations. This lack of transparency means that law enforcement can use facial recognition as a lead-generation tool while keeping both the database size and error rates hidden from defendants and their attorneys.
Settlement Amounts and What They Reveal
The financial settlements paid to victims of FBI facial recognition misidentification reveal both the severity of the harm and the varying legal success rates across different jurisdictions. Robert Williams received $300,000 in his June 2024 settlement with Detroit. A man wrongly jailed for nearly a week in Jefferson Parish, Louisiana received a $200,000 settlement. While these settlements represent acknowledgment of the harm caused, they are difficult to compare because different jurisdictions, different attorneys, and different circumstances produce different outcomes.
A settlement of $200,000 for nearly a week in jail is roughly $28,000 per day, while Robert Williams’ settlement comes out to a higher per-day value, but both pale in comparison to the decades-long wrongful convictions that facial recognition errors could potentially contribute to if the initial misidentification is not caught. not all victims of facial recognition errors receive settlements of this magnitude. Some cases are still pending in court, such as Jason Killinger’s case heading toward 2026 trial. The amount of compensation available depends on the strength of the case, the jurisdiction, the quality of legal representation, and the plaintiff’s ability to prove damages. No established standard exists for how much a wrongful facial recognition arrest should cost law enforcement—each case must be litigated separately, creating significant barriers for victims who lack legal resources.
The Critical Lack of Due Process Protections
Perhaps the most alarming aspect of FBI facial recognition use is that defendants are often not even told when facial recognition has been used to identify them. In a February 2025 hearing, FBI witnesses could not confirm whether the agency informs criminal defendants when facial recognition has been used in their case. This revelation highlighted a fundamental due process violation: how can someone defend themselves against an identification technique if they don’t even know it was used? Even when defendants do learn that facial recognition was involved, they are typically denied access to critical information about the technology used against them.
Defendants are routinely unable to access the algorithm’s error rates, the source code that generated their match, the training data used to build the model, or the confidence scores assigned to their particular match. This information asymmetry violates due process rights by preventing adequate cross-examination and testing of the evidence. The ACLU and the American Bar Association have both highlighted these due process gaps, noting that a defendant cannot effectively challenge evidence they are not allowed to examine. Without access to this information, a defendant cannot demonstrate that an algorithm known to be 100 times more likely to misidentify Black people was actually wrong in their specific case.
The ACLU’s Federal Lawsuit
The American Civil Liberties Union is currently suing the FBI, the Drug Enforcement Administration (DEA), U.S. Immigration and Customs Enforcement (ICE), and U.S. Customs and Border Protection (CBP) over their facial recognition use and lack of safeguards. This federal lawsuit represents the broadest legal challenge yet to law enforcement’s use of facial recognition technology at the federal level.
The ACLU’s argument is straightforward: these agencies have deployed biased technology at massive scale without establishing adequate procedures to prevent false identifications, verify accuracy before making arrests, or protect defendants’ constitutional rights. The ACLU lawsuit is significant because it goes beyond individual wrongful arrest settlements and challenges the entire system of federal facial recognition use. If the ACLU prevails, it could force the FBI and other agencies to implement strict validation requirements for their algorithms, establish protocols to inform defendants when facial recognition is used, and provide defendants with access to algorithmic error rates and matching data. This litigation could also lead to mandatory bias testing and potential limitations on how facial recognition can be used in criminal investigations.
The Road Ahead and Calls for Algorithm Validation
The central claim in these lawsuits—that the FBI and other federal agencies used unvalidated algorithms on Black defendants—appears well-supported by evidence. The NIST study documenting the racial disparities in facial recognition accuracy is not speculative; it’s peer-reviewed research. The settled cases and ongoing trials are not hypothetical; they represent real people who were arrested based on faulty identifications. The 640 million image FBI database is not an exaggeration; it’s the agency’s own figure.
Moving forward, civil rights advocates, researchers, and some legislators are calling for mandatory validation requirements before facial recognition can be used to support criminal arrests. This would mean that before an agency could use facial recognition as investigative lead, it would need to publicly demonstrate that its algorithms meet minimum accuracy standards for all racial groups. Several states and some jurisdictions have already begun implementing such requirements. The question now is whether federal law enforcement will face similar mandates through litigation, legislation, or administrative action. The outcome will likely determine whether facial recognition becomes more or less common in criminal investigations in the coming years.