Anthropic announced on 23 September 2026 that its Claude model had helped identify a previously uncharacterised enzyme system in bacteriophage DNA, which the company said carries features “reminiscent” of the repeat structures found in CRISPR systems. According to Anthropic’s own announcement, the system, which it named array-associated reverse transcriptase (ART), was flagged by an AI agent after it combed through more than 200,000 reverse transcriptase sequences and noticed an unusual repeating DNA pattern next to an odd-looking reverse transcriptase gene, a process the company said took roughly 950 Claude agents 21 hours and 210 million tokens to complete (anthropic.com).
Anthropic’s post describes the moment of discovery in vivid terms, quoting the agent as exclaiming that the sequence “is spectacular” and resembled “a CRISPR-like… repeat array,” before the system proceeded to count repeats, measure spacing and search the literature the way a human researcher might, according to the company’s own account (anthropic.com). The company said early laboratory experiments showed the ART repeat array is expressed as distinct short RNAs, a feature it likened to how CRISPR arrays generate the RNAs that make CRISPR-Cas systems programmable. Anthropic was careful to note that the primary function of ART remains unknown and that further experiments are underway.
Independent reporting adds caveats to the claim
Coverage from outlets outside Anthropic’s own communications has repeated the company’s framing while adding notes of scientific caution. The Verge reported that despite Anthropic’s comparisons to CRISPR, it remains unclear whether the discovery will have any practical application at all, let alone the kind of transformative impact CRISPR itself has had on medicine (theverge.com). Reuters, in a report carried by WDSM, noted that the underlying reverse transcriptase had already been identified in earlier scientific studies, meaning Claude’s contribution was to notice new structural features around a previously known enzyme rather than to discover an entirely new molecule from scratch (wdsm710.com). A separate summary from Interesting Engineering stated plainly that researchers caution ART has not been shown to function like CRISPR, and that its biological role remains unknown pending further experiments on the reverse transcriptase, the accessory protein and the repeat-derived RNAs (interestingengineering.com).
Wire coverage from Channel News Asia confirmed the basic mechanics of Anthropic’s claim, describing how Claude analysed large DNA databases and flagged the unusual RT-associated array, while stopping short of endorsing Anthropic’s framing that the finding is comparable in significance to the original discovery of CRISPR (channelnewsasia.com).
How the outlets framed it
Anthropic’s own announcement leans heavily on the CRISPR comparison, describing ART as a system with a “pattern reminiscent of CRISPR” and framing the discovery as evidence that Claude can autonomously drive biological research. The Verge and Interesting Engineering, by contrast, foreground the unresolved question of function, stressing that researchers have not shown ART actually behaves like CRISPR and that its practical value is unknown. Reuters’ wire report, carried by WDSM, adds the detail that the underlying enzyme was already known to science, which tempers the novelty claim. The gap between Anthropic’s promotional language and the more measured caveats in independent coverage illustrates how a company with a direct commercial interest in demonstrating its AI’s scientific capability may describe a preliminary finding in more dramatic terms than outside observers are willing to accept.
No outlet reviewed for this article carried on-the-record comment from Washington University microbiologist Kevin Blake, whose criticism of the CRISPR comparison was reported elsewhere; that specific rebuttal could not be independently verified against a live, reachable source at the time of writing and is therefore not repeated here as fact.
