AI analysis challenges authenticity of Rubens painting with mixed results

Artificial intelligence’s role in art authentication is emerging as a contentious frontier in the centuries-old practice of determining artistic provenance. A recent AI analysis of “The Bath of Diana,” long considered a copy of a lost Rubens masterpiece, has sparked fresh debate about the painting’s authenticity and highlights the complex relationship between traditional connoisseurship and emerging technological approaches to art verification.

The big picture: Swiss authentication startup Art Recognition claims AI analysis indicates parts of “The Bath of Diana” may have been painted by Peter Paul Rubens himself, contradicting the long-held belief that the work is merely a copy.

• The company’s AI model analyzed 29 distinct patches of the painting, finding varying probabilities of authenticity across different sections of the artwork.
• This mixed-authentication result aligns with Rubens’s known practice of employing workshop assistants, suggesting potential collaboration between the master and his studio.

By the numbers: Art Recognition’s AI authentication delivered distinctly varied results across different sections of the painting.

• 10 patches were deemed authentic with over 80% probability, while 8 patches showed 60-80% probability of authenticity.
• 7 sections yielded inconclusive results, and 4 patches—including the central figure of Diana—were classified as inauthentic.

What experts are saying: Leading Rubens authority Nils Büttner, chairman of the Centrum Rubenianum in Antwerp, has challenged the AI attribution despite being generally supportive of technological authentication tools.

• Büttner cited a 2016 condition report and his own 2023 inspection that identified several technical inconsistencies with Rubens’s known working methods.
• Technical issues include the use of a reddish canvas primer atypical of Rubens’s technique, inconsistent underdrawing, and inferior painting quality.

Behind the limitations: Art Recognition’s CEO Carina Popovic acknowledged potential dataset limitations in their analysis.

• The authentication was conducted before a collaboration that subsequently improved their Rubens dataset, potentially affecting the reliability of these specific results.

Why this matters: The case demonstrates both the promise and limitations of applying artificial intelligence to art authentication, a field traditionally dominated by connoisseurship and technical analysis.

• Successful AI authentication systems will require close collaboration between technologists and art specialists to ensure comprehensive datasets and appropriate applications.
• The debate highlights how emerging technologies are reshaping traditional practices in the art world, creating both opportunities and controversies.