Fingerprints have always been considered the holy grail of forensic identification — and beyond.
Then came artificial intelligence.
---
Fingerprints remain a central tool for biometric user authentication to this day.
From authorizing payments on a phone and gaining entry to buildings, to cracking criminal cases in court.
The reliance on fingerprints as a unique identifier stems from the belief that every fingerprint is one of a kind in the universe, and that each of our fingers differs absolutely from the others.
---
A recent study trained an AI model on a government database of 60,000 fingerprints.
Surprisingly, the model can now link two different fingerprints belonging to the same person with 77% accuracy.
The implications could be far-reaching: it would become possible to connect individuals to crime scenes even when the fingerprint found there came from a different finger.
It also enables the identification of similar fingerprint patterns across multiple crime scenes, thereby linking them to a single perpetrator.
---
This capability rests on a new approach to fingerprint identification.
In traditional identification systems, fingerprint recognition relies on dry parameters such as the length and position of ridge lines.
The AI model, by contrast, also examines patterns such as the curvature and angles of those ridges — and in doing so, succeeds in detecting a characteristic pattern shared across all of a person's fingerprints.
---
After considerable doubt and extensive review, the researchers decided to publish the study, which is now undergoing deeper scrutiny.
In the next phase, the model will likely be trained on a larger database to improve its accuracy, and will be incorporated as an auxiliary tool in forensic examinations.
It is important to note that the research does not challenge the uniqueness of individual fingerprints — a fingerprint therefore remains an effective biometric identifier for payments and similar purposes.
The study's primary impact lies in unsettling the foundational assumptions of forensic identification, which has always focused on finding an exact match to a single finger rather than also leveraging patterns to match additional fingers.