Facial Recognition Fails in Real World, Study Reveals

The Discrepancy Between Lab Performance and Real-World Accuracy

Facial recognition technology has become a common tool in various sectors, including law enforcement and security. However, its real-world performance often falls short of the high accuracy scores reported in controlled environments. This gap between laboratory results and actual application raises serious concerns about the reliability and fairness of these systems.

Academics from the University of Oxford, including Teo Canmetin, Juliette Zaccour, and Luc Rocher, have highlighted this issue in a post on the Tech Policy Press website. They argue that public failures of facial recognition systems demonstrate that their real-world effectiveness does not match the impressive statistics from lab tests.

The US National Institute of Standards and Technology (NIST) conducts the Facial Recognition Technology Evaluation (FRTE), which is often used to justify the deployment of AI systems. However, the Oxford researchers point out several flaws in this benchmarking process. First, the evaluation does not account for real-world conditions such as blurred or obscured images. Second, the datasets used are too small, increasing the likelihood of misidentification. Third, these datasets fail to represent the diverse demographics found in the real world.

These shortcomings have led to significant real-world failures. For instance, a Detroit man was wrongfully arrested in 2020 due to flawed facial recognition. A study by the University of Essex also revealed that a live facial recognition system in London only accurately identified eight out of 42 faces. These incidents underscore the risks associated with relying on inaccurate systems.

According to the Oxford researchers, the latest models report accuracy rates as high as 99.95 percent. While these numbers suggest exceptional performance, they do not reflect the challenges faced in real-world scenarios. "Out of context, these numbers suggest that facial recognition has progressed to be extremely accurate. But there's a problem: these near-perfect numbers fail to reflect reality. Facial recognition appears to be significantly less accurate in real-world settings," they wrote.

A May 2025 research paper from the University of Pennsylvania further supports these findings. The study, titled "Accuracy and Fairness of Facial Recognition Technology in Low-Quality Police Images: An Experiment With Synthetic Faces," found that facial recognition technology (FRT) performance degrades under poor image conditions. Factors like blur, pose variation, and reduced resolution negatively impact accuracy. Moreover, the degradation is not evenly distributed across demographic groups, leading to higher false positive and false negative rates for individuals from marginalized race and gender groups.

Despite these issues, algorithmic accuracy remains only one aspect of the broader concerns surrounding FRT. A 2023 US Government Accountability Office report revealed that many law enforcement agencies use FRT without adequate training or civil rights policies. This lack of oversight can lead to misuse and potential violations of individual rights.

The consequences of this lack of regulation are evident in the Algorithmic Justice League's "Comply To Fly?" report. It found that the US Transportation Security Administration (TSA) uses FRT without adequately informing travelers. Many passengers are unaware they can opt out of FRT scans, and two-thirds report experiencing hostile treatment when attempting to do so.

NIST did not immediately respond to requests for comment. However, the US government standards body recently published guidelines on detecting face morphing, a technique that combines multiple faces to create a composite, potentially to deceive FRT-based authentication systems.

A February 2024 report by the Innocence Project, authored by Alyxaundria Sanford, echoed the concerns raised by the Oxford researchers. It noted at least seven confirmed cases of misidentification due to FRT, with six involving Black individuals who were wrongly accused. The Electronic Frontier Foundation added two more names to this list—Christopher Galtin and Jason Vernau—who were wrongly arrested following flawed FRT identification.

Advocacy groups argue that regardless of its accuracy, facial recognition technology poses too great a risk for police use and should be banned. As the debate over the ethical implications of FRT continues, it is clear that the technology’s real-world performance must be scrutinized more closely to ensure fairness and prevent harm.