Updated
Updated · Financial Times · Sep 29
Anthropic's Claude Mirrors Unpublished Enzyme Research, Raising Snoop-and-Scoop Fears at 1 AI Lab
Updated
Updated · Financial Times · Sep 29

Anthropic's Claude Mirrors Unpublished Enzyme Research, Raising Snoop-and-Scoop Fears at 1 AI Lab

1 articles · Updated · Financial Times · Sep 29

Summary

  • Anthropic last week said Claude had independently identified a new enzyme system, but University of Copenhagen biologist Mario Rodríguez Mestre says the result closely matches his unpublished work using Claude.
  • Anthropic said it knew of no previously published paper with the same finding, argued Claude's key contribution was linking the reverse transcriptase to a larger system, and said the model was not trained on user transcripts.
  • Mestre says some of the same unusual genes from the same phages appeared in both efforts, though he acknowledges the overlap could be coincidence and says he has no evidence Anthropic directly used his unpublished material.
  • The dispute has sharpened calls for AI-assisted research to disclose what data, tools and human guidance were used, especially as confidential grant reviews and peer-review materials could expose novel ideas before scientists publish them.
  • A similar complaint recently hit OpenAI over work on Navier-Stokes, widening concern that AI firms could claim 'autonomous' breakthroughs that mirror researchers' unpublished paths.

Insights

When Claude and a scientist reach the same unpublished enzyme finding, who deserves discovery credit—and what proof could settle it?
Can an AI truly make an autonomous biological discovery if humans supply the prompts, data access, and final validation?
Will science need mandatory AI provenance logs before companies can credibly claim their models discovered something new?