AI analysis finds £71,000 painting dismissed as copy is a Caravaggio (www.theguardian.com)

A Swiss art-authentication firm, Art Recognition, working with researchers including Liverpool University, used AI-based forensic analysis to conclude with 85.7% probability that a painting long dismissed as a copy—The Lute Player from Badminton House—was painted by Caravaggio. The model returned a “strong match” against verified Caravaggios and also flagged the Wildenstein version as likely inauthentic. The finding is supported by non-AI evidence: provenance linking the work to 17th‑century inventories, a match to Giovanni Baglione’s detailed description, and technical scrutiny of lute depictions by a leading luthier. Art Recognition’s director noted that scores above 80% are “very high,” and proponents say the result corroborates a holistic forensic case for reattribution. For the AI/ML community this is a high-profile example of computational attribution influencing scholarship and the art market: a machine‑learning system produced probabilistic, reproducible evidence that challenges long‑standing expert consensus at institutions like Sotheby’s and the Met. The case illustrates both the power and limits of AI in cultural heritage—models can quantify stylistic similarity and surface features, but outcomes remain intertwined with provenance research, technical art history, and expert debate. The painting’s potential reattribution would have major market and curatorial implications, and further peer-reviewed publication and public discussion (including an upcoming podcast and academic paper) are expected to test and refine the claim.