Few paintings have intrigued art historians as thoroughly as The Polish Rider (around 1655), one of Rembrandt’s most famous works. Housed in New York’s Frick Collection, the canvas, its subject and its attribution to Rembrandt have been a subject of debate for decades. It shows a young man in military attire, riding through a dark, mysterious landscape, his identity and his expression as enigmatic as the artist’s role in its creation. Now, a leading art authentication firm, the Zurich-based Art Recognition, has run an analysis of the painting using its software powered by artificial intelligence (AI) to see if it can provide further insights into the extent of Rembrandt’s involvement in the canvas.
The Polish Rider, like many of the Dutch master’s works, became caught up in the febrile atmosphere created by the Rembrandt Research Project (RRP) during its most prescriptive phase, when its committee of connoisseurs at one stage reduced its list of works by the artist from 650 to 265 works. In that atmosphere a throwaway remark in a 1984 book review by the RRP’s then chairman Josua Bruyn—that The Polish Rider “shows striking affinities, to say the least, with [Rembrandt’s assistant Willem] Drost’s early work”, a comment made in anticipation of considering the painting fully in the RRP’s series of Rembrandt Corpus volumes (something that would wait until 2010)—led some Rembrandt lovers to believe that the RRP had removed the painting from his accepted oeuvre. It had not.
A ‘minor disaster’
The publication of the article, and the myths that built from it, were a “minor disaster”, the Rembrandt scholar Ernst van de Wetering wrote in 2014 in the concluding volume of the RRP Corpus catalogue raisonné, “all the more because other Rembrandt specialists now also began to doubt the painting’s authenticity”, while focusing on the picture’s mysterious subject matter—was it a portrait of an actual Polish aristocrat or a stock Dutch take on the Polish cavalryman at a time when Poland’s war with Sweden was of great interest to its Dutch trading partners?—and its restored and reconstructed condition. “Once sown,” Wetering wrote, “these doubts took root, even though in 1997 it was reported reliably that the RRP considered the painting to be an authentic Rembrandt.”
Was the painting a portrait of an actual Polish aristocrat or a stock Dutch take on the Polish cavalryman at a time when Poland’s war with Sweden was of great interest to its Dutch trading partners?
The painting had been acquired in 1910 by the US industrialist Henry Clay Frick—it had been in Polish aristocratic and royal collections since the 1790s—as a particularly magnificent Rembrandt, one that has been a star attraction at the Frick for more than a century. In the wake of the 1984 Bruyn article, some experts suggested that parts of the painting, if not the entire composition, were executed by one of Rembrandt’s students, with Willem Drost the primary candidate. Others focused on the painting’s stylistic anomalies. The rider’s face, while finely rendered, lacks the complexity of expression typically found in Rembrandt’s portraits. The landscape, too, has been criticised as lacking the atmospheric depth that characterises his other works. Some have noted that the brushwork in certain areas appears inconsistent with Rembrandt’s usual technique, particularly in the treatment of the horse and the rider’s garments.
Defenders of the Rembrandt attribution argue that the painting’s strengths outweigh its oddities. They point to the rider’s pose, the subtle interplay of light on his armour, and the emotional tension that pervades the scene—qualities that align with Rembrandt’s mastery of drama and human expression. It is also possible, they argue, that the painting is a collaboration between Rembrandt and his studio. Others proposed that Rembrandt began the work but left it unfinished, to be completed by a follower. The apparently unfinished elements include the horse’s hindquarters, tail and legs, the rider’s hands and patches of landscape.
Many of the painter’s doubters may have been inspired by the curious treatment of the horse’s feet on the ground. But, as Wetering pointed out in the final Corpus volume: “these two feet were in fact painted by a later restorer, and not altogether competently. They had to be re-painted because a strip had been cut from the bottom of the painting and replaced with a new strip of linen”. (The bottom four inches of the painting had been damaged in the 19th century, requiring the addition of the strip and new painting on it.) In volume five of the Corpus (published in 2010, by which time the RRP had modified its approach to authentication, moving away from the old committee model), the painting is subjected to extensive analysis, including the work of a 19th-century restorer on the horse’s hooves on the new strip of linen, and the work done in the same area by William Suhr of the Frick in 1950. The RRP’s opinion of the painting in its 2014 volume reads “Rembrandt The Polish Rider (partly unfinished, locally completed by later hand)”.
The power of AI
Art Recognition has shown the most promise and enjoyed the most success of any firm harnessing the power of AI to provide new tools for art analysis. Founded in 2019 by Carina Popovici, a former physicist with a PhD in computational science, Art Recognition has developed AI algorithms to analyse and assess the authenticity of paintings, offering a more data-driven approach than traditional connoisseurship. Provenance research of course contributes to authentication work, as does forensic testing (using infrared spectroscopy or x-rays) in a conservation lab.
By combining her technical expertise with a passion for art, Popovici aimed to create an AI tool that could aid art historians, collectors and museums in making informed decisions. She is careful to stress that AI systems like hers will never replace the expert human eye, but can be a valuable additional analytical tool. And Art Recognition’s work can be done remotely as it is based on high-resolution digital images.
AI learns the main characteristics of the artist, and also to distinguish genuine art from non-authentic examplesCarina Popovici, Art Recognition
Art Recognition has already been used to support claims in the authentication of works by masters such as Raphael and Anthony van Dyck. Its system is designed to detect subtle patterns in brushwork and composition, providing a supplement to human judgement. Popovici explains how it works: “We train the AI on photographs of all known works of art by a chosen artist, as well as negative examples such as images of known fakes, art created by followers, pupils, school, circle, etc. and even digital art produced by a generative AI in the style of that artist. From all these images, the AI learns the main characteristics of the artist, and also to distinguish genuine art from non-authentic examples. The most important feature is the brushstroke, but the AI also learns other features: edge locations to distinguish object structures from one another, chromatics, high-level elements of composition. Once the training has been completed, the programme compares the learned characteristics with those identified on the image of a work of art. Based on this comparison, the AI returns a probability for the authenticity of the new art piece.”
Twice restored in the 19th century
The Polish Rider is trickier than most to analyse because of the new strip of linen added to its base and because the painting’s physical condition is not optimal; it was restored in 1833 and 1877, and cleaned in 1898 and 1950. Art Recognition software was trained in what to look for, and what not to be fooled by, and once high-resolution images had been run through the AI model, the results proved very informative.
The output indicates that the hand of restorers is visible in the hooves and lower parts of the horse’s front-left and back-right legs, as well as in the ground on which they stand. This is the result of 19th-century interventions by overzealous restorers. None of these elements is by Rembrandt, the software analysis showed.
The remaining sections of the painting are by Rembrandt, according to the recognition model, ranging from a 69% to 83% probability. Art Recognition analyses section by section and considers any probability over 60% to be convincing evidence that the section is by the artist in question. The difference between the sections that have a 69% probability of being by Rembrandt and those with an 83% probability is, according to this author, down to the involvement of the artist’s studio, typical in its era, in helping the master complete a painting.
Art Recognition’s analysis of the painting of the horse’s feet and the extent of Rembrandt’s involvement in the canvas aligns with Wetering’s judgements in the 2010 and 2014 RRP Corpus volumes. A spokesperson for the Frick declined to comment on Art Recognition’s findings, citing museum policy.
The history of the painting’s reception, and how technology has been used to address it, feel like one more reason—combined with the re-opening of the Frick’s restored historic Fifth Avenue home in April 2025—for visitors to commune with the collection and with Rembrandt’s mysterious pale rider.
• Noah Charney is an art historian and author of the novel The Art Thief (2007)