Anthropic's "AI-Made Biological Discovery" Claim Collides With the Researchers Who Say They Got There First

Anthropic's "AI-Made Biological Discovery" Claim Collides With the Researchers Who Say They Got There First

Update, October 3, 2026. The dispute over Anthropic's claim that Claude agents discovered a new enzyme system has sharpened around a specific fear: that the model's "discovery" may have drawn on the very researcher's own unpublished work, fed to Claude in his chats.

The researcher is Mario Rodríguez Mestre of the University of Copenhagen, who told The Scientist and the New York Times that he had studied the same reverse-transcriptase system — which his team nicknamed "jumbotrons" after the jumbo phages they came from — for four years, identifying the specific "array-associated RT" (ART) enzyme a year ago. He first worked on these RTs in 2022 while consulting for ReNAgade Therapeutics, which lists him as a co-inventor on related patents. He used Claude extensively throughout, sharing dissertation and manuscript drafts with the model.

His central concern is no longer just priority but provenance: "My concern is that this information was used to train future versions of the models¹," Mestre said. And on whether the system was even new: "So, for me, the most important question is not 'Were ARTs already known?' They were."

Anthropic's position, per its announcement and The Scientist's account: its research team identified the ART "in raw data from a 2022 paper in which Mestre was not involved" (Zhang et al., Microorganisms 2022, on jumbo phage SA1). But the company "did not share any of the exact prompts its research team had fed into Claude" — and given that opacity, Mestre "couldn't say definitively that the AI had sampled his work," while calling the similarity between the two projects "striking."

The episode lands as AI-for-science claims face a broader provenance reckoning — The Scientist notes it "is just the latest to strike AI's forays into research, following a mathematician's claim that AI company OpenAI had borrowed from his work to crack a major unsolved math problem" — and it complicates the celebratory framing of Anthropic's in-house wet lab and its related commercial push (see also Novo (fka Novo Nordisk) Partners with Anthropic to Put Claude Science Inside Drug Discovery). The unresolved questions: what exactly was in Anthropic's prompts, whether Mestre's chat data could have reached the agents, and whether ARTs meet the bar of a discovery at all.