Anthropic says its biology lab has already found something big

Anthropic announced that a wet biology laboratory it operates in the San Francisco Bay Area has identified a previously unreported enzyme system encoded in bacteriophage DNA that the company says can perform DNA operations such as cutting, copying and pasting in ways analogous to CRISPR. The discovery was disclosed after Anthropic last week confirmed the existence of the lab and said it uses AI models to guide physical experiments. The company said Claude—their large language model—played a major role in the computational discovery process. According to Anthropic, Claude performed the bulk of the work over about 21 hours using roughly 950 autonomous agents and consuming approximately 210 million tokens. Anthropic’s CEO Dario Amodei acknowledged that a Stanford team had already described a related system and framed the new result as building on prior work. Anthropic was careful to stress that the hands-on laboratory work was carried out exclusively by human scientists. The firm said the Bay Area facility resembles a standard molecular biology lab, operates at biosafety levels BSL-1 and BSL-2, and does not handle pathogens that can infect humans. The company also noted it is not currently allowing Claude to autonomously control lab equipment, though Amodei said that could be possible in the future with appropriate safeguards. The announcement comes amid broader debate in the AI community about the risks posed by increasingly capable models in biological research. Several AI leaders, including Amodei, have recently called for slowing development and implementing safety testing, and some employees have publicly raised extreme-risk concerns. Anthropic’s disclosure underscores both the accelerating use of AI in life-sciences research—where groups at Stanford, UC San Francisco and Google have published related work—and the steps companies are taking to retain human oversight while deploying these tools.