A man rides through a darkening landscape, his face lit and unreadable, his horse's hind legs strangely flat and lifeless against the ground. He has puzzled historians for over a century, not because of who he is, though nobody is quite sure of that either, but because of who painted him.
The Polish Rider hangs in the Frick Collection in New York and has done since 1910, when the American industrialist Henry Clay Frick bought it believing it to be a Rembrandt of exceptional quality. For most of the twentieth century, nobody seriously argued otherwise. Then, in 1984, the art historian Josua Bruyn, chairman of the formidable Rembrandt Research Project, wrote that the painting showed 'striking affinities, to say the least' with the early work of Willem Drost, one of Rembrandt's assistants.
That single sentence detonated. Rumours spread that the Rembrandt Research Project had quietly struck the painting from the master's official catalogue. It had not, but the doubt stuck anyway, the way doubts do. Scholars began picking at the painting's face, which lacked the psychological density of Rembrandt's other portraits, and at the landscape, which seemed to want depth and not find it. The brushwork on the horse and the rider's clothing struck some as simply wrong, a hand imitating Rembrandt rather than being him.
The Rembrandt scholar Ernst van de Wetering, writing in 2014 in the final volume of the Rembrandt Research Project's catalogue raisonné, called the whole episode a 'small disaster'. The truth, he explained, was far less dramatic than the rumour: the project had reaffirmed the painting as an authentic Rembrandt as far back as 1997, a fact that somehow never caught up with the myth. What had actually happened to the painting was almost mundane by comparison. The bottom ten centimetres of the canvas had been damaged in the nineteenth century, and a strip of new linen had been grafted on and repainted, not altogether skilfully, to patch the horse's hind legs and hooves. It was restored in 1833 and 1877, cleaned in 1898 and again in 1950. The 'incompetent' patch job, not Rembrandt's hand, was likely what generations of eyes had been reacting to.
Into this decades old argument has now stepped an algorithm. Art Recognition, a Zurich company founded in 2019 by the physicist turned art authenticator Carina Popovici, has run the Polish Rider through its artificial intelligence model to measure how much of the painting is genuinely Rembrandt's. The system, as reported by Noah Charney in Il Giornale dell'Arte, is trained on photographs of an artist's confirmed body of work alongside deliberate negative examples: known forgeries, works by students and followers, even AI generated pastiches in the artist's style. From these, the model learns to recognise brushwork, colour handling and compositional habits too fine grained for the eye alone, then returns a probability score for a disputed work.
The Polish Rider was, by Popovici's own account, an unusually hard case, precisely because of its patchwork history: the grafted canvas strip, the repeated cleanings, the nineteenth century repainting layered over the original. Even so, the model's analysis pointed to exactly where the restorers' hands had intervened: the hooves and lower portions of the horse, the same areas Van de Wetering had already identified from documentary research alone.
What is quietly remarkable here is not that a machine settled an argument. It has not, and Popovici herself is careful to say her tool cannot replace the trained eye, only sharpen it. What is remarkable is that decades of connoisseurship, archival digging and one very human bout of academic panic have now been checked, area by area, brushstroke by brushstroke, against a model that has never seen a Rembrandt except as pixels. The machine and the scholars agree. The horse's legs were always the problem. Rembrandt just was not the one who painted them.
