Anthropic Says Claude Autonomously Discovered a New Enzyme System Comparable to Crispr
Nearly 950 Claude agents analyzed 210 million tokens over 21 hours, flagging a repeating DNA pattern that led researchers to a previously uncharacterized enzyme system in bacteriophages. Anthropic says its function and practical value remain unknown.
Anthropic announced in September 2026 that its AI model Claude has “autonomously discovered” a new enzyme system with similarities to the machinery behind Crispr, the powerful gene-editing technology. The finding is the first result from the company’s newly launched wet lab.
According to Anthropic, Claude identified the enzyme system by searching through a massive database of DNA sequences. Over 21 hours, nearly 950 Claude agents processed 210 million tokens before one flagged an unusual repeating pattern in the dataset for human review. Subsequent analysis and lab testing revealed “a previously uncharacterized enzyme system found in bacteriophages” — a type of virus that infects bacteria. The company said its scientists’ involvement was limited to the initial prompt and follow-up lab work.
Anthropic said it is still working to understand what the enzyme system does, but chose to share the finding early, citing a desire to demonstrate Claude’s capabilities and provide the broader scientific community with visibility into its research. The company acknowledged the announcement is premature and that it remains unclear whether the discovery will have any practical applications, let alone prove as consequential as Crispr.
The disclosure comes as Anthropic prepares to go public and looks to attract more scientists to its lab, with plans to expand into areas such as drug discovery. The company is among a growing number of AI firms turning to scientific research to demonstrate the value of their more capable models. OpenAI, the source notes, has separately been working through advanced mathematical problems, drawing scrutiny over its approach to academic norms.
For readers following AI development, the result may suggest that large language models can contribute to biological research at a meaningful level — though the practical significance of this particular finding remains to be determined.