Anthropic confirms Bay Area lab for Claude biology experiments

The Bay Area lab did not appear on its own. Anthropic introduced the Model Hardware Standard on August 27 as a research preview, connecting agents to instruments including microscopes, liquid handlers, and robotic arms. On September 17, the company reported optimizing more than 30 biomolecular models in under four weeks, and it released the resulting code on GitHub.

The company's own reporting suggests where the gaps remain. A protein design test found that about $150 in GPU computing and AI usage yielded predicted binding scores comparable to earlier campaigns allowed up to $10,000 per target. But Anthropic's MHS university case study identifies handoffs between research stages as an obstacle to automating the full cycle, and in a pilot with QuEra, Claude could not troubleshoot physical hardware failures and sometimes paused overnight for human confirmation.

How fully Anthropic has connected Claude, the Model Hardware Standard, and its own lab is not yet clear. The reading that seems supported is that the physical lab matters less as a standalone facility than as a way to test whether measurements can feed back into the next design round — with failures, such as the designs that did not bind, becoming useful input rather than dead ends.