For more than 100 years, fingerprint identification has been treated as an unshakable pillar of forensic science. Courts, police forces, and investigators across the world have relied on one core assumption: no two fingerprints are the same, not even among the ten fingers of a single person. A new artificial intelligence study now suggests that this belief may no longer hold true.
A peer-reviewed study published in Science Advances has revealed that fingerprints taken from different fingers of the same individual can share hidden similarities that traditional forensic methods fail to detect. These similarities are invisible to the human eye but can be identified using advanced deep-learning systems, raising serious questions about how fingerprint evidence is interpreted in criminal investigations.
The research was conducted by scientists from Columbia University’s School of Engineering and the University at Buffalo. The team analysed more than 60,000 fingerprint images using an AI model designed to determine whether two prints came from the same person, even if they were taken from different fingers. The results showed consistent structural patterns across an individual’s fingerprints, challenging the long-standing idea of absolute uniqueness.
Unlike conventional fingerprint analysis, which focuses on minutiae such as ridge endings and bifurcations, the AI system examined broader features like ridge flow, curvature, and orientation. Using a deep-learning method known as contrastive learning, the neural network learned to distinguish subtle similarities and differences between fingerprint pairs. The model achieved an accuracy rate of about 77 percent when matching fingerprints from different fingers of the same person. When multiple samples were analysed together, confidence levels exceeded 99.99 percent.
To ensure the findings were reliable, researchers trained and tested the AI using multiple well-established fingerprint databases, including NIST SD300, NIST SD302, and the UB RidgeBase dataset. They carefully controlled for image quality, sensor differences, lighting conditions, and data overlap between training and testing sets. The analysis consistently showed that ridge orientation near the centre of fingerprints played a key role in identifying cross-finger similarities, while minutiae contributed far less than expected.
The implications for forensic investigations could be significant. In simulated crime-solving scenarios, the AI system dramatically reduced suspect lists. Instead of comparing all ten fingers of thousands of individuals, the model was able to narrow down potential matches by more than 90 percent. This could help investigators link partial or unclear fingerprints from different crime scenes to the same individual far more efficiently.
However, the researchers also urge caution. While the AI showed consistent performance across gender and racial groups, accuracy was highest when training and testing datasets shared similar demographic profiles. This highlights the need for larger, more diverse fingerprint databases before such systems are used in real-world legal settings.
Importantly, the study does not claim that fingerprints are identical between people. Rather, it shows that fingerprints from the same individual are not as independent as forensic science has long assumed. This distinction matters, especially in courtrooms where fingerprint evidence is often treated as definitive proof.
The findings mark a turning point in biometric science, suggesting that artificial intelligence can uncover patterns long hidden within familiar data. As AI continues to evolve, the study raises urgent questions about how forensic standards should adapt, and whether one of criminal justice’s most trusted tools needs to be re-examined in the age of intelligent machines.