Anthropic’s Biology Discovery Under Scrutiny Over Source of Research Data

Allegations arise after a researcher claims AI used unpublished data in a “discovery” of a protein linked to bacterial gene-editing systems.

Image credit:©iStock.com, ILYA SEDYKH

AI company Anthropic recently reported that its Claude model identified a novel enzyme variant, previously undocumented in scientific literature. However, the breakthrough has come under scrutiny after a University of Copenhagen researcher alleged the AI may have utilized unshared data from ongoing research to reach its conclusion.

Genetic Discovery in Question

<

The contested finding centers on a tandem repeat array in reverse transcriptase (RT) proteins, which Anthropic termed “array-associated RT” (ART). The company announced this discovery followed an intensive search across 20,000 candidate proteins, claiming ART paralleled enzyme systems integral to CRISPR gene-editing mechanisms. Mario Rodríguez Mestre, the researcher who initially identified the system, disputed these conclusions, stating his team had encountered the enzyme family four years prior during work at ReNAgade Therapeutics.

<

“The question isn’t whether these proteins were known,” Mestre explained to The New York Times. “They were. My concern is whether Claude accessed unpublished data from my research process.” Mestre disclosed he and colleagues had identified ART one year into their four-year study, naming the proteins “jumbotrons” due to their presence in jumbo phages. He emphasized using Claude extensively for analysis, including sharing manuscript drafts with the AI model.

Anthropic Responds to Claims

<

The company issued a statement to The Times denying user data involvement: “Claude was not trained on user transcripts, and our molecular biology team has no access to such data.” Anthropic further noted its discovery derived from a 2022 paper in which Mestre was not a participant, though it declined to release specific prompt inputs used in the investigation.

Mestre’s patent filings co-signed with ReNAgade Therapeutics in 2022 corroborate his prior work on RT systems. Despite the controversy, he acknowledged the “striking similarity” between his findings and Anthropic’s report. The dispute mirrors recent AI research debates, including a mathematician’s claim of OpenAI leveraging his work for an unsolvable problem resolution.

Source link