Anthropic's Claude finds a novel CRISPR-like enzyme system in new biology lab

Anthropic said on 23 September 2026 that it has set up a new life-sciences research group and laboratory, and that in one of its first research programmes an AI system built on Claude autonomously discovered a novel enzyme system in bacteriophage DNA with a pattern of repeats "reminiscent of CRISPR." TechCrunch confirmed the announcement the same day.

How Claude made the find

• Anthropic said it gave Claude "a prompt to search through a massive database of DNA sequences" for reverse transcriptases (RTs), enzymes that copy RNA into DNA.
• Claude agents gathered more than 200,000 RTs and picked out 3,500 candidate systems, narrowing those down to "the 20 most-compelling candidates," according to the company.
• The search took "21 hours spent searching this data by roughly 950 agents using 210 million tokens," Anthropic said, before one agent flagged a repeating DNA pattern next to an unusual reverse-transcriptase gene.
• Anthropic named the resulting system "array-associated reverse transcriptases (ART)" and said its biological function is not yet known.
• Feng Zhang, a CRISPR pioneer at MIT and the Broad Institute, said after reviewing Anthropic's pre-print that the finding is "genuinely intriguing and merits further investigation," according to Anthropic.
• Anthropic said "all of the lab work is performed by human scientists" at biosafety levels BSL-1 and BSL-2; chief executive Dario Amodei wrote on X that the discovery was made "mostly, though not entirely, by Claude," TechCrunch reported.

Why it matters

The case is one of the most concrete public examples so far of a frontier AI system credited with an original scientific finding rather than summarising existing work, though Anthropic itself says the enzyme's function remains unknown and the result still needs validation beyond its own pre-print. It also arrives days after Amodei and other AI chief executives were reported to be discussing independent safety oversight for increasingly capable models, a tension TechCrunch highlighted given Amodei has separately named bioterrorism as one of his biggest concerns about advanced AI. For universities and research institutes across the Caspian region weighing agentic AI tools for scientific work, the case is an early signal of what multi-agent systems can plausibly do in hypothesis generation, alongside the safety-governance questions that come with letting them operate largely unsupervised on biological data.