Anthropic says Claude found a novel CRISPR-like enzyme system

By Splitfeed EditorialLast updated Sep 29, 2026, 18:21 UTC

In late September 2026, Anthropic announced that Claude agents, operating inside an internal molecular biology lab the company launched earlier that year, had identified a novel enzyme system in bacteriophages with a CRISPR-like structure. The lab's model has Claude reading literature, generating hypotheses about difficult biology problems, and directing human scientists to run experiments based on those outputs. Anthropic framed Claude's identification of an array of RNA molecules and a protein associated with reverse transcriptase as an autonomous first step in biological discovery.

The announcement immediately attracted serious challenge on two fronts. First, a researcher noted that experimental validation of the finding had not been completed before the claim went public, characterising a situation where experiments were still in the queue while the PR was already live. Second, and more specifically, a scientist named Dr. Mestre stated that he and his colleagues had already made the same discovery and had in fact shared key findings about those enzymes with Anthropic. That puts Anthropic in the position of having announced as a novel AI discovery something that prior human researchers say they identified years earlier and briefed the company on directly. The deeper question the episode forces is what standard of evidence and originality should qualify as an AI scientific discovery at all.

The Anti-AI camp is the only one with published reactions at this point, and those reactions cluster around two charges: prior art and possible data misuse. Accounts including Barbara J. King, Mindy Weisberger, Sara Soueidan and Paul Kingsley all point to Dr. Mestre's claim that the systems Anthropic highlighted are ones his team had studied for years and shared with the company. Hagen Blix raises a separate concern, suggesting the results may have been achieved by training on user data, while Juspreet Singh Sandhu frames the broader pattern as academic plagiarism. The Pro-AI camp has not responded publicly yet, though stories like this typically prompt arguments that AI identifying connections across large bodies of literature is itself a meaningful contribution, regardless of whether the underlying observations existed in prior work. The Middle Ground camp has also not weighed in, but would typically be expected to push for clearer definitions of what counts as discovery credit before either dismissing or endorsing the claim.

The most consequential thing to watch is whether Anthropic responds directly to Dr. Mestre's account of sharing his findings with the company, and whether the experimental validation of the enzyme system produces results that either confirm or contradict the original announcement. Any formal peer-reviewed publication from either Anthropic's lab or Dr. Mestre's group would go some distance toward settling the priority question.