Claude finds binders for 14 of 15 targets

Anthropic says Claude produced 354 lab-confirmed protein binders from 1,320 designs against 14 of 15 targets, in a campaign that used Claude Opus 4.8 and Mythos Preview with external wet-lab validation. In Anthropic’s report, the company says overall hit rates reached 22% to 35%, versus a typical 10% to 15%, while the agent orchestrated existing structure, sequence, and folding models rather than inventing a new biology simulator. Scientific-AI builders should file this as evidence that orchestration can compress design work—not as a drug-discovery claim—and compare it with the archive’s AI drug-design evidence gate before skipping prospective validation.