Study Finds People Overconfident in Ability to Detect AI-Generated Faces

Researchers say even 'super recognizers' struggle to reliably spot fake faces created by advanced AI systems.

Published on Feb. 25, 2026

Got story updates? Submit your updates here. ›

A study from Australian researchers found that people, including those with exceptional face-identifying abilities, are becoming overconfident in their ability to distinguish real and AI-generated faces. The study showed that even 'super recognizers' could only slightly outperform average participants in accurately identifying AI-created faces, despite their high confidence in their detection skills. The researchers say the problem is that modern AI face generators have become so sophisticated that the usual visual cues used to spot fakes are no longer reliable.

Why it matters

As AI-generated faces become more realistic and indistinguishable from real ones, people's overconfidence in their ability to detect them could make them vulnerable to misinformation, fraud, and other digital deception. The findings raise concerns about the growing threat of 'deepfakes' and other AI-powered impersonation schemes.

The details

The study, published in the British Journal of Psychology, surveyed 125 people - 89 with average face-identifying abilities and 36 'super recognizers' with exceptional face recognition skills. Participants were shown images of faces and asked to determine if they were real or AI-generated. The researchers found that even the super recognizers could only slightly outperform the average participants, scoring just marginally better than chance. However, both groups expressed high confidence in their ability to spot the AI-created faces. The researchers say the problem is that modern AI face generators have become so advanced that the usual visual flaws and imperfections used to identify fakes are no longer present. Instead, the most convincing AI faces tend to appear 'too perfect' and 'too average' to be real.

  • The study was published on February 19, 2026.

The players

Dr. James Dunn

A researcher from the University of South Wales' School of Psychology and a co-author of the study.

Dr. Amy Dawel

A psychologist with the Australian National University and a co-author of the study.

Got photos? Submit your photos here. ›

What they’re saying

“People have been confident of their ability to spot a fake face. But the faces created by the most advanced face-generation systems aren't so easily detectable anymore.”

— Dr. James Dunn, Researcher, University of South Wales (British Journal of Psychology)

“Ironically, the most advanced AI faces aren't given away by what's wrong with them, but by what's too right. Rather than obvious glitches, they tend to be unusually average—highly symmetrical, well-proportioned and statistically typical.”

— Dr. Amy Dawel, Psychologist, Australian National University (British Journal of Psychology)

What’s next

The researchers say they want to learn more about the strategies used by the 'super-AI-face-detectors' they identified during the study, in the hopes of finding ways to teach those skills to the general public.

The takeaway

As AI-generated faces become increasingly realistic, people's overconfidence in their ability to detect them could leave them vulnerable to digital deception and misinformation. The findings underscore the need for greater skepticism and more advanced tools to reliably identify AI-created content.