Anthropic Says Claude Just Discovered a New CRISPR-Like Enzyme System — On Its Own

Anthropic said Wednesday that its Claude model autonomously discovered a previously uncharacterized enzyme system with structural features reminiscent of CRISPR — the first published result from the company's newly launched wet lab, and an early proof point for its bet that AI agents can accelerate fundamental biology research.

How Claude found it

Anthropic gave Claude a single high-level prompt: search a massive database of DNA sequences for interesting new examples of reverse transcriptases, enzymes that copy RNA into DNA. From there, a swarm of roughly 950 Claude agents worked for 21 hours, processing 210 million tokens, gathering more than 200,000 known reverse transcriptases, and narrowing the field to 3,500 candidate systems before flagging 20 for detailed human review. Anthropic said its scientists' direct involvement was limited to that initial prompt and the physical lab work that followed — the database search and candidate-narrowing were handled by the agents themselves.

What it actually found

One agent spotted an unusual repeating DNA pattern sitting next to a known reverse transcriptase in bacteriophage DNA — the viruses that infect bacteria. Anthropic's wet lab scientists followed up and confirmed the array produces distinct short RNAs, and the company has named the system array-associated reverse transcriptases, or ART. The underlying reverse transcriptase itself had been identified in prior research; what Claude appears to have been first to notice is the full system, including the associated non-coding DNA array and an accessory protein alongside it.

Reminiscent of CRISPR, but not yet a gene-editing tool

The repeat architecture around ART echoes the DNA repeats that first tipped scientists off to CRISPR decades ago, which is what makes the comparison notable. But Anthropic has been explicit that ART's actual biological function is still unknown, and the company has not published a peer-reviewed paper, disclosed the underlying sequence, or reported biochemical experiments establishing what the system does. Repeat-associated genes in phages can serve several different roles — replication, recombination, host manipulation, or even blocking bacterial immune defenses — so structural resemblance to CRISPR is a lead, not a conclusion.

CRISPR pioneer Feng Zhang, a professor at MIT and the Broad Institute who reviewed Anthropic's pre-print, called it "an exciting example of how AI agents can contribute to biological discovery," while noting the finding merits further investigation.

Part of a broader biology push

The announcement lands against the backdrop of Anthropic's escalating investment in AI-driven biology. The company acquired stealth biotech startup Coefficient Bio for roughly $400 million earlier this year, folding its small team of computational-biology researchers into Anthropic's health and life-sciences group, and has built out partnerships with pharmaceutical companies including Novo Nordisk, Sanofi, and AbbVie. Anthropic confirmed on September 18 that it operates a wet lab in the San Francisco Bay Area, built specifically to test Claude's computational predictions against real biological experiments — the facility this new enzyme discovery ran through.