Anthropic reported on September 23, 2026 that Claude agents helped identify an enzyme system with CRISPR-like repeats, called array-associated reverse transcriptases, or ART. The company describes computational discovery followed by laboratory work. It explicitly says the system’s primary function is still unknown.
The announcement is an early biological result, not an approved therapy or proof of a new gene-editing platform. The similarity to CRISPR is a reason to investigate, not a conclusion that ART already has CRISPR’s demonstrated uses.
What Claude identified
Anthropic says the work found a reverse-transcriptase system associated with repeat sequences and an accessory protein in bacteriophages. The underlying enzyme had appeared in prior research; the claimed contribution is recognizing the system’s defining features and investigating their relationship.
That distinction is important in scientific reporting. Finding a new relationship in known data can be a meaningful discovery without requiring every component to be new. The evidence should specify what was previously characterized, what is newly observed, and which functional explanation remains a hypothesis.
Laboratory follow-up changes the evidence
The source describes experiments showing that the repeat array is expressed as distinct short RNAs, with further characterization underway. It says human scientists perform the laboratory work. A model-generated candidate and an experimentally supported observation are different stages.
For researchers considering a similar approach, keep those stages separate in records and metrics. Count candidates proposed, candidates rejected by review, and candidates that survive experiments. A high volume of hypotheses is useful only if the screening process identifies results worth testing.
Why the unknown function belongs in the headline’s interpretation
A structural resemblance can suggest a direction for research, but it does not establish mechanism, programmability, safety, or therapeutic usefulness. These questions require further experiments and outside scrutiny.
Readers should therefore resist treating “CRISPR-like” as a synonym for “a replacement for CRISPR.” The reported result can advance biological understanding while practical applications remain undetermined. This is a case where a precise limitation makes the story more informative rather than less significant.
What scientific teams can learn operationally
The workflow combines broad computational search with specialist filtering and experiments. A useful implementation preserves dataset provenance, candidate reports, literature comparisons, and the reasons a scientist chose a candidate. That makes it possible to reproduce and challenge the conclusion.
Anthropic’s Life Sciences Verification Program is a separate access mechanism for approved research. The discovery announcement does not establish that every ordinary account can reproduce the campaign or undertake every described research task.
Nerova’s assessment is that ART is a substantive example of AI contributing to hypothesis generation and anomaly detection. The scientific contribution should be evaluated through the biological evidence, while the function and eventual applications remain open questions.