AI analysis challenges the authenticity of two Jan van Eyck paintings at major museums

7 February 2026

Saint Francis of Assisi Receiving the Stigmata, on display in the Philadelphia Museum of Art. A near-identical painting is in the Royal Museums of Turin. Photograph: incamerastock/Alamy

An artificial intelligence analysis of two small Renaissance paintings traditionally attributed to Jan van Eyck has sent ripples through the art world by calling into question who actually painted them and prompting a broader conversation about how technology is reshaping the study of Old Masters.

The works in question, two versions of Saint Francis of Assisi Receiving the Stigmata one held by the Philadelphia Museum of Art and the other by the Royal Museums of Turin in Italy have been part of the canon of van Eyck’s oeuvre for decades, regarded as rare surviving examples of a 15th-century master known for his meticulous detail and shimmering naturalism. Yet new machine-learning tests conducted by Swiss company Art Recognition found that neither panel shows the characteristic brushstrokes consistent with works firmly attributed to van Eyck, raising the possibility that both could be products of his workshop or later followers rather than the artist himself.

Jan van Eyck, a Flemish painter active in the early decades of the Northern Renaissance, is celebrated for his groundbreaking use of oil paint and astonishingly precise technique that made surfaces glow with realism. Only a small number of works are universally accepted as autograph fewer than 20 and each is studied with intense scrutiny by scholars and curators.

The disputed Saint Francis panels, both executed around 1428–1432 and nearly identical in composition, depict the saint kneeling before a rugged landscape as he receives the stigmata, a legendary moment in his life. Over the years art historians have debated their authorship, with some suggesting that differences in technique and condition might indicate one is a copy or workshop piece, but until now they have remained attributed to van Eyck or his circle.

The new analysis employed algorithms trained to recognize subtle brushwork patterns and other microscopic features drawn from high-resolution imagery of confirmed van Eyck paintings. According to Art Recognition’s findings, the Philadelphia version had a 91% “negative” score for authorship by van Eyck, while the Turin panel registered an 86% negative probability. Such results are significant because AI-based brushstroke analysis introduces a quantitative dimension to art authentication, a field traditionally dominated by connoisseurship, historical documentation and expert judgment. The same tools gave a high positive likelihood to unquestioned van Eyck works such as The Arnolfini Portrait, which adds weight to the contrasting results for the two Saint Francis pictures.

The findings have stirred debate among art historians, curators and technologists alike. Some scholars view the results as supporting earlier doubts that both paintings might be products of van Eyck’s workshop, executed under his supervision but not necessarily by his hand, a common practice in Renaissance studios. Others caution that AI algorithms, while powerful, have limitations, especially when applied to works centuries old that have undergone restoration, varnish buildup and other surface changes that can confound digital analysis. The debate highlights how technological innovation can challenge longstanding assumptions and compel scholars to re-examine their approaches to attribution.

Experts who have weighed in note that the corpus of securely attributed van Eyck paintings is small, which complicates AI training and statistical confidence. Unlike datasets in fields such as natural science or contemporary image recognition, art historical analysis must grapple with variations in technique, material degradation and the fact that Renaissance artists often worked with apprentices and assistants. In van Eyck’s era, the concept of a solitary artistic genius was less rigid; workshops frequently produced multiple versions of popular subjects, and attribution could be fluid. As a result, the idea that AI might conclusively determine authorship based on brushstroke alone remains controversial.

The museums in Philadelphia and Turin have not publicly confirmed whether they will accept the AI results or conduct further scientific testing. In the past, institutions have responded cautiously to new authentication claims, recognizing that re-evaluating attributions can have significant implications for collection interpretation, exhibition planning and even insurance valuations and market perceptions. Some art historians and museum professionals emphasize that AI findings should be integrated with traditional connoisseurship and technical analysis such as infrared reflectography, pigment dating and examination of underdrawings to build a more holistic understanding of a work’s origins.

Indeed, the unfolding discussion touches on deeper questions about the role of technology in cultural heritage. Proponents of AI analysis argue that tools like machine learning can offer fresh insights, flag potential misattributions and expand the analytical toolkit available to scholars. Critics counter that the subtleties of artistic creation, particularly in Old Master paintings where surface features can be altered by centuries of restoration, require nuanced interpretation that AI alone cannot provide. The challenge lies in developing collaborative frameworks in which computational techniques and human expertise inform each other rather than compete.

Beyond technical debates, the controversy has captivated a broader public fascinated by the intersection of art and technology. Headlines about AI potentially uncovering “fakes” in major museum collections evoke questions about authenticity, value and the stories we tell about cultural icons. For many, the idea that works long attributed to a master like van Eyck might be reclassified invites reflection on how historical narratives are constructed and how evolving tools can reshape our understanding of the past.

In the coming months, the conversation is likely to deepen as additional analyses are debated among experts and possibly published in peer-reviewed forums. Whether the Saint Francis panels are ultimately confirmed as workshop pieces, re-attributed entirely or subjected to new investigative methods, the episode underscores an ongoing evolution in how art history is practiced. AI may not replace the discerning eye of the connoisseur, but as this case illustrates, it is increasingly part of the dialogue, challenging conventions and prompting a reexamination of some of the greatest treasures of Western art.