Art authentication ultimately has to survive contact with the market.
A report may persuade scholars. A technical analysis may strengthen an attribution. Yet an auction presents a different test: will a reputable house accept the work, will collectors bid, and will anyone buy it?
In November 2024, three works analysed by Art Recognition went under the hammer at Germann Auction House in Zurich. All three sold.
The artists involved could hardly have offered a broader test: Louise Bourgeois, Marianne von Werefkin and Mimmo Paladino. Each work arrived with a different set of questions, and Art Recognition’s analysis played a different role in each case.
The Louise Bourgeois ink drawing, dated 1945, already had supporting provenance. Here, the AI analysis added another piece of evidence to an attribution that rested on existing documentation. The work sold for CHF 28,000, approximately $31,600.
Mimmo Paladino’s Lo Spirito della Foresta, from 1981, showed that image-based analysis also has a place in the contemporary market. Questions surrounding authorship, studio practice and market confidence do not end with the historical avant-garde. The Paladino sold for CHF 19,000, approximately $21,500.
The most revealing case was the watercolour attributed to Marianne von Werefkin.
Unlike the Bourgeois, it lacked strong provenance. Much of the case therefore depended on the work itself: its composition, handling and visual relationship to accepted paintings by Werefkin. Art Recognition analysed the watercolour using models trained on authenticated examples of her work and found that the image supported the attribution.
The market responded enthusiastically. The Werefkin sold for CHF 15,000, approximately $17,000. This was around $9,300 above its high estimate, making it the strongest result of the three.
The figures deserve some context. Neither the Bourgeois nor the Paladino reached its low estimate. Their importance lies elsewhere. Buyers accepted all three works, and Germann was willing to offer them with Art Recognition’s findings included in the wider body of evidence. The sale placed AI-assisted authentication inside an actual commercial transaction, where reputations and money were at stake.
Fabio Sidler, a partner at Germann Auctions, saw the sale as the start of a broader development: “We have many reasons to believe that our partnership will encourage other art dealers to include AI authentication in their expertise.”
The Germann sale also showed that there is no single formula for using AI in authentication. In one case, it reinforced a provenance-backed attribution. In another, it supported a contemporary work entering the market. In the Werefkin case, the visual analysis carried far greater weight because the documentary trail was thin.
That range is precisely what makes the sale useful. AI authentication did not arrive with one dramatic discovery or one disputed masterpiece. It appeared in three ordinary market situations, each with its own level of evidence, uncertainty and risk.
All three works found buyers. The Werefkin surpassed expectations. And a method once discussed mainly in theoretical terms took its place in the auction room.