Artificial intelligence executives are betting that large language models can transform biology as dramatically as they have changed coding, software development and mathematics. Anthropic’s latest attempt to prove that case has instead triggered skepticism and controversy among scientists.

Anthropic, the company behind Claude, said its biology research lab used AI agents to identify an unusual DNA pattern in the genetic code of viruses. The company described the finding as a “novel enzyme system.”

The signature resembles the microbial system used in gene editing tools such as CRISPR Cas-9, which can modify the DNA of living organisms. CRISPR technology is common in laboratories and earned its inventors the 2020 Nobel Prize in chemistry.

Anthropic chief executive Dario Amodei presented the result as scientifically promising. “Today we announced the Claude-led discovery of a molecular machine that we suspect could represent a new gene editing mechanism,” he wrote on X.

The company said the discovery required “only high-level direction from our scientists.” Shares of companies developing gene editing technology fell shortly after Anthropic announced the finding.

Scientists who were not involved in the research offered a more restrained assessment. They described the result as preliminary and incremental rather than evidence of a major biological breakthrough.

“These are very interesting preliminary results, but at this point they don’t demonstrate gene editing or provide a mechanistic picture of exactly what’s happening,” said Aaron Engelhart, an associate professor at the University of Minnesota’s Department of Genetics, Cell Biology, and Development.

Anthropic’s researchers have not established what the newly identified viral sequences do or whether they could support practical applications comparable to CRISPR tools. The research also has not appeared in a peer reviewed scientific journal.

Some experts nevertheless said the pattern deserves further study. Anthropic quoted Feng Zhang, a professor at MIT and the Broad Institute and a pioneer of CRISPR genome editing, describing the work as “genuinely intriguing.”

Days after the announcement, Mario Rodríguez Mestre said his unpublished doctoral research at the University of Copenhagen described the same viral signatures. Mestre said he had used Claude extensively during that research.

“It is essentially the same finding. This is not simply a case of two groups studying related protein families or similar biological systems,” Mestre told CNN by email.

Mestre said he had uploaded unpublished research to his private Claude account, including dissertation drafts, analyses and material about the systems. He said he could not demonstrate that Anthropic deliberately took the work and considered greater prior scientific knowledge and human direction another possible explanation.

Anthropic did not respond to an emailed request for comment. Ilya Finkelstein, a professor in molecular biosciences at the University of Texas at Austin, said scientists may think twice about putting unpublished findings into such tools.

Anthropic said researchers prompted Claude to search a massive DNA sequence database for interesting examples of reverse transcriptase, an enzyme that copies RNA into DNA. A total of 950 AI agents searched for 21 hours, identified 3,500 reverse transcriptases and narrowed them to 20 compelling candidates.

One agent detected repeating DNA sequences beside the gene for an unusual reverse transcriptase. Anthropic’s scientists then analyzed and tested the finding, describing it as a “previously uncharacterized” system in bacteriophages, viruses that attack bacteria.

Gustavo Sudre, a professor of genomic neuroimaging and artificial intelligence at King’s College London, said the broad assignment showed greater scientific autonomy. Finkelstein, however, said AI excels at large searches and repetitive work when experts guide it, rather than at imaginative breakthroughs.

Finkelstein noted that lead author Dr. Peter Yoon previously worked in the laboratory of Jennifer Doudna, one of the architects of CRISPR Cas-9. He also said four of the paper’s six authors are senior molecular biologists and domain experts.

Anthropic has launched Claude Science, while OpenAI, Microsoft and Google DeepMind have introduced their own scientific tools. Such systems can search specialized datasets, but their reliability remains tied to the quality and focus of the information used to train them.

The crucial next step is physical laboratory experimentation to determine what the pattern does and whether it has any biotechnology use. Scientists said robotics is not yet ready to perform all the delicate experimental work, while human judgment remains essential for deciding which scientific questions and results matter.