Anthropic says Claude helped identify a CRISPR-like enzyme system, opening a new test for AI-driven biology

Anthropic’s claim that Claude helped identify a new enzyme system marks a notable step in the effort to turn large AI models into scientific collaborators rather than only software assistants. Reuters reported that the company said Claude analyzed large DNA databases and found an unusual system built around reverse transcriptase, an enzyme that copies RNA into DNA. Anthropic has named the system array-associated reverse transcriptases, or ART, and said it has features reminiscent of the repeating DNA patterns seen in CRISPR systems.

The announcement is significant because biology is one of the fields where AI companies hope models can move beyond summarizing papers and begin generating useful research leads. Large biological databases contain patterns that may be difficult for humans to inspect manually, especially when meaningful signals are scattered across genomes, protein families and non-coding DNA regions. If a model can search those spaces and propose hypotheses that later hold up in laboratory work, AI could accelerate discovery in ways that look different from ordinary text generation.

Reuters reported that the underlying reverse transcriptase had been identified in earlier studies, but Anthropic said Claude appeared to be the first to recognize important features of the broader system. Those features include an array of non-coding DNA sequences and an additional protein whose function is not yet known. That distinction matters. The claim is not that Claude created a biological mechanism from nothing, but that it connected pieces in existing data into a system scientists had not previously characterized in that way.

The result arrives as Anthropic expands into life sciences. Reuters recently reported that the company had quietly established a wet lab in the San Francisco Bay Area, a sign that its biology ambitions are moving beyond purely computational analysis. Wet-lab validation is crucial because AI-generated scientific leads can be plausible, elegant and wrong. In biology, a useful pattern must survive experiments, replication and careful interpretation before it becomes a tool or therapy. Anthropic’s own account frames the work as early research rather than a finished medical application, which makes the boundary between discovery, validation and product promise especially important for readers to understand.

The ART finding also lands in a sensitive regulatory environment. AI models that can help discover biological systems may benefit medicine, agriculture and basic science, but the same capability raises concerns about misuse. Systems that reason across DNA databases, enzymes and cellular mechanisms could eventually help design helpful therapies or harmful biological tools. The more convincing the scientific claims become, the more pressure companies will face to explain what safeguards govern model access, lab work and publication decisions.

For researchers, the important question is whether Claude’s contribution can be independently verified and extended. If outside scientists can reproduce the analysis, test ART experimentally and map what the system does, the discovery will become more than a corporate milestone. It would be an early example of AI helping to identify a biological mechanism hidden in public data. If not, it will still serve as a warning that AI science claims need the same scrutiny as any other research claim, perhaps more because the surrounding commercial incentives are so intense.