Scientist says Anthropic's AI “discovery” may have relied on his unpublished research

By isabelle // 2026-09-29

Mastodon

Parler

Gab

Copy

A Copenhagen biologist disputes Anthropic's claim that its Claude AI independently discovered new enzymes.
Mario Rodríguez Mestre says he found the enzymes in 2022 and shared unpublished research with Claude.
Anthropic says Claude was not trained on user transcripts and its biology team had no access to them.
Mestre admits he has no proof but is moving his work to other AI models.
The dispute comes as Anthropic eyes an IPO that could value it at $2 trillion.

A University of Copenhagen researcher is disputing Anthropic's claim that its Claude AI independently found a new biological system with possible gene-editing uses. Mario Rodríguez Mestre told the New York Times over the weekend that he first identified the enzymes, which his team calls "jumbotrons", in 2022, years before Anthropic announced its finding. At issue is whether the AI truly made the discovery itself or was steered toward it by unpublished research Mestre's group had shared with Claude.

Biologist says he found enzymes four years before Anthropic

Mestre, a computational biologist, spotted the enzymes while hunting for new reverse transcriptases in jumbo phages, oversized viruses that prey on bacteria. He briefly consulted for ReNAgade Therapeutics, which filed a 2023 patent on several reverse transcriptases, including those he found, and he is named as a co-inventor. The group has yet to publish, but for three years, Mestre and his colleagues have used Claude for coding, drafting manuscripts and other routine work. Mestre said he gave Claude drafts of his dissertation and a jumbotron manuscript along the way. Records he shared with the Times, including Slack messages, appear to show the AI was deeply involved in the research.

Anthropic says its AI agents combed billions of gene sequences

Last week, Anthropic said AI agents in its new biology lab had found a family of enzymes it named array-associated reverse transcriptases, or ARTs, and that its AI had taken "the first steps of a biological discovery on its own." Company scientists gave the agents five known reverse transcriptases—enzymes that copy RNA into DNA—and pointed them at databases holding billions of gene sequences from oceans, soil and other environments. Anthropic has not disclosed its exact prompts. The agents wrote their own search tools and flagged candidate genes in jumbo phages sitting beside arrays of RNA genes. Anthropic also said the AI independently tracked down raw data from a 2022 experiment measuring RNA in a microbe infected by a jumbo phage. The data showed high levels of ART-linked RNA, which the company said supported its theory of how the enzymes work.

Researcher questions whether unpublished work fed into AI models

Mestre said his team analyzed that same dataset about a year ago, used it to pinpoint the jumbotrons' RNA genes and shared the unpublished results with Claude. "My concern is that this information was used to train future versions of the models," Mestre said. "I think it's important to raise this possibility." He acknowledged he has no proof and that coincidences happen in a field where many researchers chase similar leads. Anthropic rejected the suggestion. In a Sunday statement, the company said it knew of no previously published work describing the ART system, that Claude was not trained on user transcripts, and that its molecular biology team has no access to them. Seth Shipman, a bioengineer at the Gladstone Institutes and the University of California, San Francisco, said many specialists already knew of these systems. His own lab relies on Anthropic's tools, and he said the announcement has worried colleagues. "I've heard from a bunch of people in our field who say, 'Uh oh, are they training on my prompts?'" he said. Profluent, an AI biotech lab, told CNET that this reverse transcriptase had been seen before; what Anthropic added was the neighboring array and a partner protein.

Dispute echoes OpenAI math controversy as Anthropic eyes IPO

A similar controversy arose this month involving OpenAI, after mathematicians said they had been using OpenAI's models while working on a solution to a longstanding problem that the company later claimed its technology had solved. OpenAI denied training on one mathematician's conversations. Anthropic, which is expected to go public at a valuation as high as $2 trillion, has made biology a centerpiece of its pitch. At the United Nations Security Council on Wednesday, Chief Executive Dario Amodei said AI could bring "profound progress in the near future on diseases that have plagued humanity for millennia." Mestre, who started using the research-focused Claude Science just last month, is now shutting down his Claude projects and moving to other AI models. Until Anthropic shows exactly how its agents reached their finding, including the prompts it has kept private, researchers are being asked to take a company with trillion-dollar ambitions at its word. For scientists who hand their unpublished work to these tools every day, that is a lot to ask. Sources for this article include: YourNews.com NYTimes.com CNet.com