Anthropic Claims Claude Found a New Enzyme System; CRISPR Researchers Call It Routine Genome Mining

Anthropic says its large language model Claude identified an enzyme system that had not been described before, after searching public DNA sequence databases and carrying out most of the analysis without human direction. The company framed the result as evidence that AI models can contribute to scientific discovery on their own. The claim drew immediate pushback from CRISPR researchers, who described the workflow as routine genome mining rather than a breakthrough.

Genome mining is a long-established computational practice: researchers scan sequenced genomes for genes that resemble known enzymes, then infer what the encoded proteins might do. The approach has turned up countless enzyme families over the past two decades, often with automated pipelines doing the heavy lifting. Several scientists who work on CRISPR systems said the steps Anthropic described — database search, sequence comparison, candidate selection — map closely onto that standard process.

The dispute is less about whether the enzyme system is real than about what counts as a discovery. Finding a new sequence in a database is not the same as showing what it does in a living cell, and functional characterization typically requires laboratory experiments. Without that validation, researchers said, the finding sits in the same category as many uncharacterized gene clusters already flagged by conventional tools.

Anthropic's broader argument is that models like Claude can chain together searches, read papers, and reason across results in ways that speed up early-stage biology. That pitch has become common across AI labs, which have pointed to model-assisted work in materials science and protein structure as proof of concept. But each new claim now faces a skeptical audience of domain experts who can check the method against decades of existing practice.

For the field, the episode highlights a widening gap between how AI companies describe model outputs and how biologists assign credit. If AI systems are going to be credited with discoveries, researchers say, the bar should include experimental confirmation and a clear account of what the model did that existing pipelines could not. Until then, claims of autonomous discovery are likely to keep colliding with the unglamorous reality of genome mining.

What to watch next is whether Anthropic or an outside lab publishes the sequence, the model's full reasoning trace, and any wet-lab test of the enzyme's activity. That would let independent groups judge whether Claude surfaced something genuinely new or simply sped up a search that was already routine. The answer will shape how much weight the broader AI-for-science claims carry.