Anthropic's AI 'first discovery' claim draws biologist pushback

Anthropic says its in-house molecular biology lab, staffed by around 950 Claude agents running for roughly 21 hours, surfaced a previously uncatalogued repeating pattern around a known enzyme, and called the pattern "reminiscent" of what led to CRISPR. Working biologists are not buying the framing, and that gap between the press release and the lab bench is the story MIT Technology Review uses to ask when an AI result actually counts as a discovery.

The sharpest pushback came from biologist Lucas Harrington in a post that went viral and was endorsed by the chair and CEO of Eli Lilly. "Finding a weird cluster of genes and repeats is often the easy part," Harrington wrote. "The hard part, and where the real discoveries come from, is figuring out what the system actually does." His counsel to the labs was blunt: "set the bar high now, so that when an AI actually discovers a fundamentally new biological mechanism, everyone appreciates how big a deal it is."

A second complication came from Mario Rodríguez Mestre, a biologist at the University of Copenhagen, who says his own team had already identified the same pattern, and who publicly questioned whether Anthropic's agents learned about it from his earlier conversations with Claude. The accusation sits unresolved in the piece, but it drags training-data provenance into what Anthropic framed as a clean first discovery. Two researchers we follow in our Who's Who directory passed the piece around as it moved through the biology-adjacent crowd.

Author James O'Donnell's point is less about the specific pattern than the industry habit around it: casting agents as independent discoverers rather than tools, in parallel with OpenAI's claim that its agents solved a million-dollar mathematics problem. He argues the posture undermines recognition of the genuine progress AI is making in science and makes outsiders skeptical of real breakthroughs when they come.