Anthropic Claims Claude Agents Found Uncharacterized Virus Enzyme System in Biolab Test

Artificial intelligence developer Anthropic says its Claude model has autonomously discovered a novel enzyme system in bacteriophages, drawing early comparisons to the mechanisms underpinning CRISPR gene editing. As reported by The Verge , the finding represents the first published output from Anthropic's newly established wet laboratory.

According to Anthropic, the computational search operated largely without manual intervention. Company researchers supplied the initial prompt and handled subsequent physical laboratory verification, while Claude parsed a database of DNA sequences independently. Over 21 hours, approximately 950 Claude agents processed 210 million tokens before a single agent identified an unusual repeating pattern and flagged it for human review.

Follow-up wet-lab testing confirmed the sequence corresponds to a previously uncharacterized enzyme system inside bacteriophages, which are viruses that infect bacteria. Anthropic acknowledged that its scientists are still working to determine the biological function and mechanics of the enzyme system, noting that it shared the findings early to demonstrate Claude's capabilities and offer visibility into its biological research efforts.

Whether the discovery will yield practical applications or approach the utility of CRISPR remains unproven. Anthropic's disclosure comes as the company prepares for a potential public offering and aims to recruit scientists to expand its laboratory operations into commercial fields such as drug discovery. The initiative reflects a broader industry trend among AI developers, including OpenAI, attempting to validate advanced models by applying them directly to complex academic and scientific domains.