Anthropic says Claude agents found a previously uncharacterized reverse-transcriptase system with repeating DNA structures, which human scientists then tested in a laboratory.

Anthropic’s life-sciences research group announced array-associated reverse transcriptases, or ARTs, on 23 September. The company says Claude searched large collections of DNA sequences, selected an unusual candidate and noticed an adjacent array of repeating DNA. The system resembles CRISPR in layout, but its biological function is still unknown.

Why it matters

Most AI-science claims concern benchmark questions or literature summaries. This project connected computational search to a new biological candidate and wet-lab experiments. If the finding survives independent scrutiny, it shows a general language-model agent contributing to the earliest stage of discovery: noticing a pattern that specialists had not characterized as a system.

Anthropic says roughly 950 Claude agents spent 21 hours and 210 million tokens examining reverse transcriptases, enzymes that copy RNA into DNA. They gathered more than 200,000 examples, identified 3,500 candidate systems and narrowed these to 20 reports for human review. These numbers describe Anthropic’s internal workflow and have not been independently audited.

The underlying reverse transcriptase had appeared in earlier studies. Anthropic’s claimed novelty is the connection between that enzyme, a neighboring accessory protein and a long array of evenly spaced non-coding DNA repeats. Initial experiments found that the array produces distinct short RNAs. That combination led the team to define ART as a previously uncharacterized system.

The CRISPR comparison needs restraint. CRISPR arrays store sequences that help guide programmable DNA-targeting machinery. ART has a repeat array and short RNAs, but Anthropic says it does not yet know the system’s primary function. Similar structure is a reason to investigate, not evidence that ART already cuts, copies or edits DNA in a programmable way.

Human work and open questions

Claude did not run the laboratory autonomously. Anthropic says human scientists performed all physical experiments in its Bay Area laboratory, which handles lower biosafety levels and no pathogens capable of infecting humans. The agents searched data, generated hypotheses and wrote candidate reports; people selected and tested the result.

Feng Zhang, a CRISPR researcher at the Massachusetts Institute of Technology and Broad Institute, reviewed the preprint and called the RNA-repeat association intriguing and worthy of further investigation. His comment supports scientific interest, but it is not independent replication.

The result comes from Anthropic’s own laboratory and is published as an early technical report rather than a peer-reviewed paper. The company has a commercial interest in demonstrating Claude’s scientific value. Independent groups need the sequences, analysis pipeline and experimental protocols to test novelty and function.

Attribution also matters. A discovery pipeline includes previous sequence deposits, earlier descriptions of the enzyme, software, model training data and human judgment. Saying “Claude discovered” is a convenient shorthand, but the evidence describes a human-agent system in which the model surfaced a candidate and scientists recognized, tested and named it.

The next decisive information will be a functional mechanism. Researchers need to determine what ART does in bacteriophages, whether its RNAs guide a target and whether the system can be programmed. Until then, the finding is a promising biological observation, not a new gene-editing tool.

Verification

  1. VERIFIED AS ANTHROPIC’S RESEARCH CLAIM — Anthropic announced ARTs on 23 September 2026 and released a technical preprint. Primary source: https://www.anthropic.com/news/claude-discovers-novel-enzyme-system
  2. VERIFIED AS VENDOR-REPORTED WORKFLOW — Anthropic reports about 950 agents, 21 hours and 210 million tokens. Primary source: the Anthropic post above.
  3. VERIFIED AS REPORTED LAB EVIDENCE — The system includes a reverse transcriptase, accessory protein and repeat array that produces short RNAs. Primary source: the Anthropic post and linked technical report.
  4. VERIFIED — Anthropic says the system’s primary function remains unknown and all physical lab work was performed by humans. Primary source: the Anthropic post above.
  5. UNVERIFIED INDEPENDENTLY — No external replication or peer-reviewed confirmation was located during this review.

Glossary candidates

  • Reverse transcriptase: An enzyme that copies RNA into DNA.
  • Bacteriophage: A virus that infects bacteria.
  • CRISPR: A microbial defense system adapted for programmable gene editing.

Cold-reader sentence: Anthropic says Claude helped identify a new repeat-associated enzyme system, but humans performed the experiments and its biological function remains unknown.