Anthropic reported that its Claude models designed protein binders that worked in independent wet-lab testing for 14 of 15 biological targets, achieving hit rates of 22 to 35 percent compared with the 10 to 15 percent typical for human-led design campaigns. Two contract laboratories, Adaptyv Bio and Twist Bioscience, synthesized and physically tested the AI-generated designs without modification. Anthropic also demonstrated Claude analyzing chemical compounds in under 25 minutes, versus 30 minutes to an hour normally required. The company said it restricted protein-design capabilities on its most powerful model due to dual-use concerns about potential bioweapon development.
