AI re-attribution suggests Holbein’s ‘unidentified woman’ is Anne Boleyn

TL;DR:

• Researchers at the University of Bradford used a computer-vision model to compare every image in the Hans Holbein corpus and argue two sketches in the royal collection have been mislabelled since the 1700s.
• The “Windsor sketch”, long thought to depict Anne Boleyn, may actually portray her mother Elizabeth Howard; the so-called “Unidentified Woman” may be Boleyn herself.
• Independent scholar Karen Davies estimates fewer than 15% of the 80-plus Holbein corpus images have contemporary documentary verification.

The work pairs Davies, whose study of mislabelling in the corpus appeared in March, with Prof Hassan Ugail, director of the Centre of Visual Computing at Bradford. Ugail’s model previously attributed a long-debated painting to Raphael. Applied to Holbein, the system clustered images by visual similarity into a matrix; the “Unidentified Woman” sat with other Boleyn-Howard family images, while the Windsor sketch grouped more closely with portraits of Elizabeth Howard.

Context

The wider attribution problem is not new. The Holbein corpus has long been known for inconsistent labelling — a portrait of Boleyn’s cousin Henry Howard, for example, turned out to depict his father. Davies’s red-hair-versus-darker-complexion argument adds documentary weight to the AI clustering result; the team frames it as opening debate rather than settling the question. A Royal Collection Trust spokesperson said the identity of the unnamed sitter had “long been the subject of debate” and welcomed further discussion.

For UK heritage institutions, the case sits in a growing pattern of computer vision reshaping art-historical attribution work. The National Gallery, Tate and Royal Collection Trust have all published computer-vision research projects in recent years, and AI attribution claims now sit alongside dendrochronology, pigment analysis and provenance research as a tool curators must engage with — even if the conclusions invite challenge. Earlier in 2026, the Hever “Rose” portrait of Boleyn was reanalysed by historians who argued the Elizabethan artist deliberately rebutted contemporary claims she had a sixth finger.

Looking forward

Davies has explicitly framed the result as a question reopened, not closed. The wider implication for UK museums and the royal collection is the direction of travel: as AI corpus-matching tools mature, conservative attributions on tens of thousands of works will face systematic re-examination. That will mean curators, conservators and provenance researchers spending more time engaging with model outputs, contesting them where the documentary record disagrees, and developing standards for how strong an AI cluster signal must be before a permanent re-attribution is justified.