The Zurich-based Germann Auction House made history last month by selling the first ever work of art solely authenticated using AI. Germann have worked with a Swiss AI authentication company, Art Recognition, in what has been described as a “pivotal test case”.

According to their website, Art Recognition’s technology “applies machine learning and computer vision techniques to determine the authenticity of artworks”. Carina Popovici, the CEO and cofounder of Art Recognition, suggested that if the probability percentage was above 95%, it would not be necessary to consult an art expert to confirm the attribution. If it was less than 80%, however, she suggested that an expert might be necessary.

Art Recognition’s software was used on three works of art for sale at Germann, although two of the three – a drawing by Louise Bourgeois (1911-2010) and a collage by Mimmo Paladino (b. 1948) – also had accompanying expert research to support the authentication. One work, however, was authenticated without any additional supporting information. This was a watercolour by Russian artist Marianne von Werefkin (1860-1938), an artist who was important in the German Expressionism movement. The watercolour sold for 15,000 CHF (around £13,400), well over its high estimate.

Popovici said that as this work “lacked any prior evidence of authenticity […] this made it a pivotal test case to gauge how the market would respond to the idea of selling an artwork authenticated solely by A.I. As we have seen, the outcome was a resounding success.”

But not everyone has been so positive about the use of AI in making art attributions. Leading British art historian Bendor Grosvenor has spoken on the topic, saying that while “AI will play an increasingly important role in helping us to recognize who painted what, and when […] the track record of AI attributions is patchy, to say the least.” There have been two occasions recently in which AI attributions have proved “patchy”: in 2021, Art Recognition announced that their algorithm found that Peter Paul Rubens’s Samson and Delilah at the National Gallery had a 91% chance of being a fake. But this claim was unsupported, and the museum quickly dismissed it. Similarly, last year, two different AI algorithms came to different conclusions as to whether a painting was by Raphael, pointing to the inaccuracies in these new technologies.

It remains to be seen if more auction houses will turn to AI authentication companies in the future. A spokesperson from Christie’s told ARTnews that they are considering how AI technologies might “enhance our productivity and efficiency”.