Anthropic says Claude autonomously found a CRISPR-like enzyme system in its first wet lab result

Anthropic's first wet lab result

Anthropic says its AI model Claude has autonomously identified a new enzyme system that resembles the biological machinery behind CRISPR gene editing. According to The Verge, the finding is the first output from Anthropic's newly opened wet lab, and it lands at a moment when the company is testing how useful Claude can be for real scientific work as it prepares to go public.

The company says Claude made the find by combing through a vast database of DNA sequences. Anthropic claims its own researchers contributed at only two points: writing the starting prompt and carrying out the physical experiments.

How the search unfolded

The search was a large-scale agent run rather than a single session. According to The Verge, over a 21-hour window close to 950 Claude agents processed roughly 210 million tokens before one of them noticed a repeating pattern in the dataset that seemed out of place and escalated it for human review. Follow-up analysis and lab testing then indicated the pattern pointed to an enzyme system in bacteriophages, viruses that infect bacteria, that had never been characterized before.

Anthropic acknowledges it does not yet understand what the enzyme system actually does. The company says it decided to share the result early anyway, arguing that doing so both demonstrates what Claude is capable of and gives the wider research community a view into its ongoing work.

An early announcement with open questions

Despite the headline comparison, The Verge notes that it remains unclear whether the discovery will lead to any practical application, let alone anything as transformative as gene editing. Anthropic itself concedes the announcement is premature.

There are also clear commercial incentives behind the timing. The Verge reports that Anthropic is working to attract more scientists to its lab and to expand into areas such as drug discovery. More broadly, AI firms have been leaning on scientific research to justify the value of their increasingly expensive frontier models. As a parallel, The Verge points to OpenAI's push through progressively harder mathematical problems, an effort it describes as showing little regard for the norms and traditions of that field.

Why it matters

Even with the caveats, this is a notable data point in the debate over whether AI can move beyond assisting scientists and start driving discoveries of its own. Anthropic's framing is that human input was limited to a prompt and bench work, while the conceptual step of noticing the odd pattern and knowing it warranted escalation came from the model. If similar results accumulate, it would strengthen the case for agentic AI as a genuine research instrument rather than a writing and coding aid.

For now, the significance rests largely on Anthropic's own account. The company says it still does not know what the enzyme system does, and as the history of CRISPR itself shows, the road from a strange biological observation to a usable technology is a long one. The strategic signal may matter more in the short term: AI companies are racing to prove their models can produce original science, and early wet-lab results like this are becoming the currency of that race.