Anthropic's Claude discovers CRISPR-like enzyme system ART
Anthropic announced that Claude discovered a new enzyme system called ART (Array-associated Reverse Transcriptase) in phage DNA, structurally similar to CRISPR. The discovery was made almost entirely by Claude: 950 agents ran for 21 hours consuming 210 million tokens, screening over 200,000 reverse transcriptases to find 20 candidates. ART contains a reverse transcriptase, a partner gene, and repeated DNA sequences transcribed into short RNAs, suggesting programmable editing potential. Its exact function remains unknown.
Key insights
ART system structure mimics CRISPR
ART consists of three components: a reverse transcriptase enzyme, a partner gene of unknown function, and a long array of evenly spaced repeated DNA sequences. This array structure is similar to CRISPR's repeat arrays that store guide RNAs. Experiments have confirmed that ART's array is also transcribed into distinct short RNAs, hinting at a similar programmable mechanism for targeting specific genomic locations.
Claude conducted the discovery process nearly independently
Researchers gave Claude an initial prompt to find interesting novel reverse transcriptases in massive DNA databases. Approximately 950 Claude agents worked for 21 hours, consuming 210 million tokens. They screened over 200,000 reverse transcriptases, narrowed to 3,500 candidates, then selected the top 20 and wrote human-readable reports. One agent read raw sequences, identified the repeat array, counted repeats, measured spacing, compared with known systems, and confirmed novelty before submitting to humans.
CRISPR pioneer Feng Zhang praised the discovery
According to the posts, Feng Zhang (MIT professor and CRISPR pioneer) reviewed the preprint and commented that this is a good example of an AI agent contributing to biological discovery, and that the array deserves further investigation. This endorsement lends weight to the significance of the finding.
Anthropic's life sciences team and wet lab
The team at Anthropic, formed in spring 2025, operates a wet lab in the Bay Area at BSL-1 and BSL-2 levels, not handling human pathogens. All experiments were conducted by human scientists. The tools used are publicly available Claude Science and Claude Code, meaning other labs could replicate the workflow. Anthropic is currently soliciting research collaboration proposals from external scientists.
Dario Amodei outlines exponential trend in AI for biology
Anthropic CEO Dario Amodei posted that he sees a pattern where AI performance in new intellectual domains goes from weak to superhuman within a few years, citing math competition performance from 2023 to 2026. He believes AI for biology is on a similar exponential trend. He notes that while biology requires experimentation, humans can collaborate with AI to run experiments and validate findings in weeks. He also highlights that accelerating fundamental biology discovery could increase the number of drug candidates in the pipeline.