In Italy, Artificial Intelligence Found a Stolen Painting on which Thieves Carefully Painted Over a Halo and a Lamb. Now the Algorithm Will Begin Hunting for the Treasures of Ukraine Looted by Russians

In October 2024, at the annual antique fair in Parma, Italian police photographed displayed objects and uploaded them to an internal database. Two hours later, the system issued a warning about a painting depicting a fair-skinned young woman in aquamarine and red attire, holding a semi-unrolled parchment scroll. As BBC reports, the next day officers confiscated the canvas for examination.

Investigators from the Command for the Protection of Cultural Heritage found that the original painting was stolen on August 11, 2004. According to an update from the Artnews portal, the canvas was stolen from a small church in the city of Sulmona (Abruzzo region).

Павялічыць

The one-meter high 17th-century painting originally depicted Saint Agnes. The thieves painted over the white lamb, transforming it into a scroll, and also erased the halo and a palm branch, turning the Christian saint into a young pagan woman.

The human eye would not be able to identify the repainted work based on a textual description from the database, but the algorithm ignored the absence of the scroll in the original. Artificial intelligence matched color shades, brushstroke geometry, and general silhouettes.

On May 20, 2025, the painting was returned to the church. The antique dealer avoided criminal liability due to the expiration of the statute of limitations, stating that the canvas came to him as an inheritance through a family business.

Павялічыць

How Stolen Art is Tracked Down

Since 1969, the Italian police have maintained a register of stolen cultural assets. This is the world's largest database, containing over one million objects. Previously, investigators had to manually search for matches by keywords, comparing modern lots with black-and-white, blurry, or overexposed photographs from the last century.

Павялічыць

In 2020, a team led by Lieutenant Colonel and IT engineer Salvatore Rapicavoli developed the Swoads system. It is capable of recognizing fragments of paintings that have been cut into pieces (a standard practice when stealing large altarpieces) and ignoring later additions or overpaintings. Since November 2023, the algorithm has helped Italians recover more than 300 major paintings and statues, as well as over 165 thousand smaller objects — coins, medals, manuscripts, and stamps — with a total value of 203 million euros (226 million dollars).

The shift of art trade to online platforms and the darknet allowed the system to operate in continuous scraping mode. In 2026, a team in Turin, thanks to automatic monitoring of auction house websites, found Filippo Palizzi's painting "The Shepherdess's Rest," which had been considered missing since June 1972.

The System will be Adapted to Search for Works of Art Stolen in Ukraine

Karina Popovich, co-founder of Art Recognition, and Timothy Carpenter, former head of the FBI's art crime unit, emphasize that AI remains only an auxiliary tool. The system requires immense human resources to verify the generated leads. The Italian unit employs about 300 people, while the staff of the analogous FBI unit in its best times numbered a maximum of 24 employees.

The technology currently struggles with three-dimensional objects: if a vase is photographed from the front for a police report, the algorithm will not recognize it in an auction photo taken from the back. The system is also unable to distinguish an original from an authorized copy. In the Interpol database, Murillo's painting "Boys Eating Grapes and Melon" is listed as stolen, although its original hangs in the Alte Pinakothek in Munich. The algorithm is looking for a copy by Murillo's pupil, stolen in Bucharest in 1998, and another copy stolen in Rome in 2009.

Criminals have simultaneously started using generative networks for digitally restoring damaged areas on stolen canvases and forging provenance documents.

Павялічыць

Lieutenant Colonel Rapicavoli admits that the program is unlikely to find works by Antonello da Messina or Caravaggio's famous "Nativity." The most high-profile thefts are carried out on direct commission, and such canvases immediately end up in private vaults or free economic zones, bypassing the open market.

Currently, Italian investigators are updating the software, adding databases with rare Ukrainian manuscripts and icons, which are being massively looted during the Russian invasion.

By 2027, the creation of a unified international identification network is planned. Similar developments are already operational in other countries: Turkey, with the help of AI, recovered a 15th-century tile from the UK; Spain found Renaissance tables stolen in 1979 and three gold coins of the Roman Empire valued at 200 thousand euros (224 thousand dollars); and researchers from Santa Clara University in the USA created the AI Provenance Assistant to search for art confiscated by the Nazis.

• 2
• 0
• 0
• 0
• 0
• 0
• 0
• 0