Claude’s ART Discovery: What Anthropic Found—and What Remains Unknown

On September 23, 2026, Anthropic introduced a life sciences research group and laboratory and described an early result: Claude agents had identified array-associated reverse transcriptases (ART), a previously uncharacterized system in bacteriophages, the viruses that infect bacteria. Its repeating DNA resembles a feature of CRISPR, but that resemblance does not establish what ART does—or whether it could become a useful biotechnology.122

How Claude’s search worked

Anthropic says roughly 950 Claude agents searched DNA sequence data over about 21 hours, using 210 million tokens. Reported accounts say the agents collected more than 200,000 reverse transcriptases, flagged approximately 3,500 candidate systems and selected 20 for detailed analysis.182437 Reverse transcriptases are enzymes that copy RNA into DNA.33

The search began with high-level direction from human scientists, then proceeded computationally without step-by-step human steering. One agent reportedly noticed an unusual repeat array beside a reverse-transcriptase gene—a clue that the gene might belong to a larger system. Humans subsequently analyzed the result and performed laboratory experiments. The reported autonomy applies to the computational search, not to the entire discovery-and-validation process.193742

What is ART, and what was new?

ART brings together three features: a reverse transcriptase, a neighboring partner gene and a long array of regularly spaced DNA repeats.2635 The repeat-array arrangement recalls CRISPR. It does not, by itself, show that ART has CRISPR’s function or can edit genes.122

The enzyme was not wholly new: earlier studies had identified the underlying reverse transcriptase in a jumbo phage. Anthropic’s reported advance was recognizing its association with the additional gene and non-coding repeat array as a broader system.3342 Initial laboratory work reportedly found that the array produces distinct short RNAs, but ART’s overall mechanism and practical applications remain unknown.3941

What has been published—and what would establish its significance?

The finding was released as a preprint in September 2026 and had not yet been peer reviewed in the available reporting. Coverage describes CRISPR researcher Feng Zhang’s reaction as intrigued, not as confirmation of ART’s function.192226 Anthropic’s group reportedly formed in spring 2026; its Bay Area laboratory has been described as operating at BSL-1 and BSL-2 without handling human pathogens.20

For now, ART is evidence that AI agents can help identify biological patterns worth testing, alongside scientists who design and carry out experiments. The stronger test of this approach will be whether independent work confirms the system’s novelty and establishes a meaningful biological function—not the agent count alone.11933