AI helps Italian police recover stolen Saint Agnese painting

AI-assisted image search helped Italian police locate a painting stolen from a church in 2004. Officers found it at an antiques fair in Parma in October 2024, despite changes made to disguise its religious imagery. The work, identified as Saint Agnese, was returned to its church in May 2025. The case shows how image matching can aid art-theft investigations, but it still depends on expert review and detailed records.

Officers from Italy’s Carabinieri Command for the Protection of Cultural Heritage photographed artworks during a routine visit to an antiques fair in Parma. They uploaded the images to a police database for comparison with records of stolen cultural objects. Within hours, the system flagged a painting showing a woman with a scroll.

The officers returned the following day and seized the work for expert examination. Investigators linked it to a painting stolen from the Virgo Potens church in Sulmona on August 11, 2004. The historic canvas originally depicted Saint Agnese, known in English as Saint Agnes, with a lamb and palm branch.

The Diocese of Sulmona-Valva confirmed the painting’s return on May 30, 2025. Its statement said someone had overpainted the lamb and altered other religious details to hide the work’s identity. The Diocese said the image had been changed to resemble a generic sibyl, a figure from classical mythology.

The police tool is called the Stolen Works of Art Detection System, or SWOADS. Italy’s Carabinieri describe it as a platform that combines information gathered from online sources with images collected during police operations. Its project description identifies machine learning as one of the technologies used to improve searches for stolen works.

In this case, the search went beyond matching labels such as “woman with a scroll.” The system compared visual patterns, including colours and the composition of the painting, with images in the stolen-art database. That let it flag similarities even though the scroll had replaced the lamb and other identifying features had changed.

“The machine doesn’t get bored.” — Lt Col Salvatore Rapicavoli, Italian Carabinieri cultural heritage protection unit. Rapicavoli told the BBC that the system can spot similarities that an officer might overlook while manually searching a large collection. The Carabinieri’s SWOADS description says the platform also gathers data from the web, deep web and social media.

An automated alert points investigators toward a possible match; it does not by itself establish that an artwork is stolen. In the Saint Agnese case, officers seized the painting and consulted art experts after the system raised suspicion. Their examination helped confirm the connection to the work reported missing from Sulmona.

Art Recognition co-founder Carina Popovici told the BBC: “The recovery still depends on investigators and legal cooperation.” She described AI as a force multiplier, rather than a substitute for investigators. Police still need staff to assess alerts, check records and pursue legal steps to secure and return objects.

Image matching depends on the quality and coverage of the records available to the system. A stolen object may be hard to identify if police never recorded it, or if its original photograph is poor. The tool also cannot necessarily distinguish an original painting from a copy or workshop version.

Three-dimensional objects create additional challenges because a sculpture or vase may look different from another angle. The BBC also noted that police forces need enough staff to follow up on leads produced by automated searches. Without that investigation, a similarity alert may not lead to a confirmed recovery.

The software also does not identify who stole Saint Agnese or who altered the painting. The BBC reported that the original theft could no longer be prosecuted because of the elapsed time. Recovering cultural property can still matter to the church and community, even when a case does not produce a prosecution.

Ultimately, the Saint Agnese case shows how AI can help police notice a stolen work hidden in plain sight. SWOADS flagged a possible match, but investigators and experts established the painting’s identity. The canvas returned to its church in 2025. Future recoveries will still depend on useful records and human follow-up.