Anthropic says nearly 950 Claude agents discovered a new enzyme system, comparable to Crispr's

Anthropic has announced the first result from its newly created wet lab: it claims that its Claude model "autonomously discovered" a new enzyme system with similarities to the molecular machinery that made the Crispr gene-editing tool possible. According to the company, the finding came about by having a large number of Claude agents comb through a massive database of DNA sequences, with human intervention limited to the initial prompt and the subsequent lab work.

The operational details Anthropic offers are striking: nearly 950 Claude agents worked for 21 hours and consumed 210 million tokens until one of them detected an unusual repetitive pattern in the dataset and flagged it for human review. Subsequent analysis and lab testing confirmed that it was "a previously uncharacterized enzyme system" found in bacteriophages, the viruses that infect bacteria.

Anthropic itself acknowledges that it still does not know exactly what function this enzyme system serves. Even so, it defended the decision to announce it now, arguing that it is important to share this kind of finding early, both to demonstrate Claude's capabilities and to give the scientific community a sense of the work they are doing.

The parallel with Crispr that the company itself invites deserves some qualification: despite the comparison, it is not clear whether this discovery will have any practical application, much less whether it will ever have the transformative impact that Crispr-based gene editing had. For now it is an "uncharacterized" system — detected, but with its real function still not understood.

The announcement itself is described as premature and comes at a strategic moment for Anthropic: the company is seeking to attract more scientists to its lab and expand into areas such as drug discovery, all while preparing to go public. Showing that Claude can generate science autonomously fits the narrative the company needs to build for investors and for a scientific community potentially skeptical about AI's real role in research.

The move is part of a broader trend: major AI companies are increasingly turning to scientific research as a proving ground to demonstrate the value of their most advanced models. By way of contrast, it is noted that OpenAI has been tackling increasingly advanced mathematical problems with little regard for the field's academic norms and traditions, suggesting a race among AI labs to claim scientific credit, each with its own approach and degree of rigor.

Taken together, the story combines a concrete technical fact — nearly a thousand agents analyzing genomic data until they hit on a pattern that lab analysis confirmed as real — with a corporate narrative of validation at a critical moment for Anthropic. The caution is twofold: on one hand, the finding itself still lacks functional characterization; on the other, the timing and tone of the announcement, explicitly compared to Crispr and presented as a demonstration of capabilities, invite reading it both as early-stage science and as a piece of strategic communication.