Italian police have recovered a 17th-century masterpiece stolen more than two decades ago after custom artificial intelligence (AI) software identified the heavily altered artwork at an art fair.
The one-metre-tall (3.3ft) oil painting depicting Saint Agnese (Sant’Agnese) was looted from the church of Virgo Potens in Sulmona, near Naples, on the night of 11 August 2004, reports BBC.
The discovery took place during a routine patrol at the annual antiquity fair in Parma, Italy, in October 2024, where officers from the Carabinieri Command for the Protection of Cultural Heritage photographed artworks on display.
Although thieves had repainted the canvas to disguise the saint as a young pagan woman, replacing her white lamb with a parchment scroll and erasing her halo and palm branch, the AI system flagged the snapshot within hours.
Lt Col Salvatore Rapicavoli of the Italian cultural heritage police unit confirmed that officers seized the painting following the alert, and it was officially returned to the church on 20 May 2025 after statute of limitations rules precluded criminal prosecution.
Conceived in 2020 by Rapicavoli, an IT engineer who joined the art crime unit in 2013, the tailor-made algorithm is named the Stolen Works of Art Detection System (Swoads).
Designed to identify hidden visual fingerprints, colour shades, rough silhouettes, and chopped-up fragments, Swoads continuously crawls physical catalogues and online marketplaces day and night.
Since coming online in November 2023, the software has assisted Italian authorities in recovering more than 300 large stolen paintings and statues, alongside over 165,000 smaller cultural items valued at more than €203m (£170m / $226m).
Italy has maintained a database of looted cultural artefacts containing over one million items since 1969, making it the largest repository of its kind in the world. In 2026, the Swoads software also enabled police in Turin to locate Filippo Palizzi’s painting Riposo Della Pastorella (The Rest of the Shepherdess), which had been missing since its theft from a modern art gallery in June 1972.
However, art authentication and law enforcement experts stress that AI serves as a force multiplier rather than a complete solution. Carina Popovici, co-founder of Art Recognition, noted that AI depends entirely on existing records of stolen works and requires human investigators to execute recoveries.
Timothy Carpenter, former head of the FBI’s Art Crime Team and founder of Argus Cultural Property Consultants, emphasised that pursuing AI-generated leads requires substantial staffing, pointing out that Italy employs around 300 art crime personnel compared to the FBI’s peak strength of 24.
Experts also highlighted technical limitations, such as difficulties recognising three-dimensional objects, challenges in distinguishing authorised or unauthorised copies from originals, such as stolen versions of Murillo’s The Melon Eaters, and the potential for criminals to exploit AI to alter images or fabricate false provenance documents.
Moreover, intellectual property attorney Steven Frank noted that AI algorithms are unlikely to recover high-profile masterworks such as Caravaggio’s Nativity or recently stolen pieces by Antonello Da Messina, as thieves rarely attempt to sell famous stolen art through public channels.
Italian authorities are currently expanding the system’s watchlists to cover vulnerable items like rare Ukrainian manuscripts and aim to join an internationally interconnected identification network by 2027.
Similar AI tracking tools are being deployed globally, including Turkey’s artefact tracking system, France’s Arte-Fact app, Germany’s Kiku project, US academic tools for identifying plundered Cambodian and Nazi-looted art, and Interpol’s global ID-Art mobile application.
