A swarm of agents using the Claude AI model analyzed genomic data to identify a molecular machine that resembles a CRISPR array in bacteriophage DNA. To find the system, which Anthropic calls ART (array-associated reverse transcriptases), 950 Claude agents spent 21 hours consuming 210 million tokens to filter 1.94 billion protein clusters down to 20 candidates. Human scientists then experimentally validated the find in a BSL1/BSL2 facility, confirming the system produces RNA molecules, though its precise function and potential as a gene editing tool remain unknown.

The discovery is the first output from Anthropic's new molecular biology lab, supporting CEO Dario Amodei's vision that AI could eventually help cure most diseases in 5 to 10 years. While Amodei believes AI for biology is on an exponential trend and may one day autonomously control lab equipment, some biologists argue the results are merely incremental. Experts in genome mining note that the technique of identifying reverse transcriptases next to repetitive DNA has been standard since 2008 and does not constitute a fundamental biological breakthrough.