Anthropic Builds Biology Lab to Test Claude in Life Sciences

Anthropic has quietly established its own biological laboratory in the San Francisco Bay Area, signaling a deeper move into life sciences beyond selling AI research tools. Life sciences head Eric Kauderer-Abrams confirmed the lab to Reuters, saying biological hypotheses still require wet-lab validation. The lab will complement external partnerships by letting Anthropic test how Claude performs in real experimental workflows. The company's latest scientific result shows the experimental approach already yielding leads. A team of about 950 Claude agents screened more than 200,000 reverse transcriptase-related DNA sequences and identified a previously unknown enzyme system with CRISPR-like repeat arrays. Researchers named the system ART, for array-associated reverse transcriptases. Laboratory experiments confirmed that the repeats can produce short RNA, but its biological function remains unresolved. The result suggests Claude can help uncover candidate biological phenomena, although human scientists still set the direction, review candidates and run validation. This fits a broader strategy. Anthropic has launched Claude for Life Sciences and Claude Science to connect existing research platforms, databases, professional models and workflows. In one protein-design study, Claude organized specialized tools to generate and screen designs; 1,320 candidates were sent to an external lab, and 354 bound target proteins, a hit rate of about 26.8%, with 14 of 15 targets producing at least one successful binder. The company also launched BioMysteryBench and worked with institutions including the Allen Institute and Howard Hughes Medical Institute to test whether Claude can analyze biological data and assist experimental design. Anthropic now argues that open-ended discovery creates new bottlenecks. AI can produce hundreds or thousands of candidate leads, but laboratories cannot test them all. To shorten the gap between computation and experiment, Anthropic is building a biological lab, hiring computational and experimental scientists, and developing Model Hardware Standard with HHMI's Janelia Research Campus to let AI agents control microscopes, liquid handlers and robotic arms. It has also optimized more than 30 open-source biomolecular models, reporting average speedups of about 4 times. The company acquired Coefficient Bio's computational drug discovery team and recruited Nobel laureate John Jumper. The company says it is not becoming a drug developer. It does not plan commercial clinical trials and aims to avoid competing with biotechnology firms. Instead, Anthropic wants Claude to serve as an AI infrastructure layer for life sciences, connecting data, models, tools and experimental capacity, with early emphasis on rare diseases, neglected diseases and difficult therapeutic targets.