Anthropic has announced that its Claude artificial intelligence model helped discover a previously unknown enzyme system with properties reminiscent of CRISPR, the gene-editing technology that transformed modern biology. The finding, disclosed on Wednesday, September 23, 2026, is the first result from Anthropic’s brand-new biology research lab — and the AI reportedly did the heavy analytical work in just 21 hours.
The company has named the system array-associated reverse transcriptases, or ART. Anthropic says Claude analyzed large DNA databases and spotted an unusual system built around a reverse transcriptase — an enzyme that copies RNA into DNA — surrounded by repeating DNA sequences that resemble patterns seen in CRISPR. (Reuters)
What Claude found: meet ART
ART is not a single molecule but a system: a reverse transcriptase, an array of evenly spaced, non-coding DNA repeats, and an accessory protein of unknown function. The repeating DNA sequences look structurally similar to the guide-RNA arrays that make CRISPR programmable, which is why Anthropic describes ART as having “properties reminiscent of mechanisms in gene-editing technology CRISPR.”
The system was found hiding in bacteriophages — viruses that infect bacteria. Anthropic says it can perform operations like cutting, copying, and pasting DNA, much like CRISPR does. Importantly, the underlying reverse transcriptase had already been identified in previous studies. What Claude appears to be the first to recognize is that these components travel together as a system with a defining signature — an array of non-coding DNA sequences plus the extra mystery protein. (TechCrunch)
How the search worked: 950 agents, 210 million tokens
The scale of the search is striking. Anthropic says Claude combed through public DNA databases using roughly 950 autonomous agents that burned through about 210 million tokens of data — in roughly 21 hours of concerted effort.
The agents narrowed down more than 200,000 reverse transcriptase sequences to about 3,500 candidate systems, then to 20 compelling finalists for detailed review, before one standout emerged. Human scientists set the initial direction and ran the follow-up laboratory experiments; the AI handled the database analysis in between, using its own judgment to decide which candidates deserved a closer look.
The wet lab behind the finding
The discovery comes just days after Reuters reported that Anthropic had quietly established a wet lab in the San Francisco Bay Area, as the company expands its ambitions in life sciences and drug-related research beyond purely computer-based work. Anthropic formed its life-sciences research group this spring, and the lab has apparently been operating for only a few months.
The speed is notable even by AI standards: a brand-new lab, run largely by AI agents under human supervision, produced its first claimed discovery within its first months of operation. It marks a bold shift for Anthropic — from building chat assistants and developer tools like Claude Opus 5.5 to doing hands-on molecular biology.
Big claims, careful caveats
Anthropic’s CEO Dario Amodei was quick to add context. On X, he acknowledged that the discovery builds on the work of others, noting that a team from Stanford had previously discovered a system “in some ways similar to the one Claude found.” He also emphasized that the finding was made “mostly, though not entirely, by Claude.”
TechCrunch notes that it will be up to the broader research community to validate how big — or how new — this discovery actually is. The underlying reverse transcriptase was already known to science, and ART’s precise biological function is still being studied. The announcement takes the form of a company blog post and a public pre-print, not a peer-reviewed paper.
That honesty matters. Anthropic is not claiming a Nobel Prize — it is claiming something perhaps more interesting: that an AI agent swarm can now do the kind of anomaly-spotting across massive datasets that used to require years of expert human intuition.
Why this matters for AI — and for science
If the finding holds up, it is a proof point for a new kind of scientific instrument: fleets of AI agents that search territory too vast for any single lab to read. Drug discovery, genomics, and materials science are all fields drowning in data — exactly the conditions where agentic AI thrives.
The announcement also lands at a telling moment. Anthropic’s leaders were just in Washington briefing the UN Security Council on AI safety, and the same AI capabilities that can hunt for enzymes can also be pointed at less benign targets — a duality regulators are only beginning to grapple with.
Expect rivals to follow fast. Google’s AlphaFold already reshaped protein science, and every frontier AI lab is now building life-sciences teams. The race is no longer just about who builds the smartest chatbot — it is about who first turns AI into a working scientist.
Conclusion
ART may or may not become the next CRISPR. But the method behind it — an AI searching 210 million tokens in 21 hours and finding something humans missed — is the real story. Anthropic has set an early, concrete benchmark for AI-driven discovery, and the scientific community now gets to check its work.
