Anthropic's Claim That Claude Discovered a CRISPR-Like System Faces a Provenance Challenge

Anthropic announced last week that its AI model Claude had autonomously discovered a novel enzyme system — dubbed ART (reverse transcriptases with tandem repeat arrays) — that resembles the Nobel Prize-winning CRISPR gene-editing technology. The findings, described in a non-peer-reviewed preprint, were presented as a landmark for AI-driven scientific research. Now a scientist is pushing back.

A researcher says the ART system bears a striking similarity to work he had previously conducted using Claude in his own conversations with the model. While Anthropic stated its team identified the system in raw data from a 2022 paper in which the researcher was not involved, the company did not share the exact prompts used to direct Claude's search, making independent verification difficult.

Anthropic responded that it was unaware of published work describing the ART system and that Claude was not trained on customer transcripts. The company's molecular biology team also lacked access to those conversations, according to its statement. The competing accounts do not definitively resolve the provenance question.

The dispute highlights a growing tension in AI-assisted research: how to attribute credit when a model may have been influenced by prior interactions that its developers cannot fully audit. Anthropic's own technical report acknowledged that Claude's search method failed to reproduce the same observation across ten additional campaigns, suggesting the discovery may owe more to fortunate initial conditions than to a reliable, repeatable process.