At an Italian antiques fair, authorities discovered a painting that was matched by the country’s cultural heritage police artificial intelligence system to a work stolen in 2004 from a church near Naples. While the piece on display appeared to depict a young woman, investigators determined that the original subject of the painting was Saint Agnes.
To obscure the work’s identity, those who altered it had overpainted key attributes: the lamb originally held by the saint was changed into a parchment scroll, and the halo and palm branch were removed. Despite these modifications, the police AI linked the fair’s image to an entry in an archival database by analyzing colour and form patterns.
How the AI system works and its results
The system, which compares millions of registered object photographs with images taken at fairs and listings posted online, began operation in 2023. According to the Italian police, since its deployment the technology has helped locate more than 300 major paintings and sculptures and over 165,000 smaller objects.
The algorithm is designed not only to match intact images but also to detect potential links in cases where works have been overpainted, cut up or rearranged, by recognising consistent visual features.
Limits and the need for human verification
The AI can only identify objects that exist in the database, and its alerts require human verification. False positives can occur due to poor-quality photos, different viewpoints of three-dimensional items, or the existence of copies. As a result, the system is a valuable investigative tool but does not replace traditional police and expert work.
According to the BBC, the recently matched painting entered police scrutiny through this process; further examinations and authentication steps are under way to establish the artwork’s precise provenance and legal status.
