Claude AI autonomously designs validated protein binders

Anthropic, a leading artificial intelligence company, has demonstrated that its Claude AI models can autonomously design protein binders with high success rates, marking a significant advance in AI-driven biological research. In a series of experiments detailed in a technical report published on August 19, 2026, the company showed that Claude designed 1,320 protein candidates across 15 different biological targets, with independent laboratory testing confirming that 354 of them successfully bound to their intended targets. This achieved an overall hit rate of 26.8%, substantially above the typical 10-15% seen in traditional protein design campaigns.

The work was conducted in collaboration with biotech firms Adaptyv Bio and Twist Bioscience, who synthesized the AI-designed protein sequences and performed wet-lab validation. The experiment utilized Anthropic's Claude models, including Mythos Preview and Opus 4.8, which operated an "agentic" workflow. After receiving an expert-written protocol, the AI system itself selected protein targets, orchestrated a suite of existing open-source design tools such as RFdiffusion and ProteinMPNN, generated candidate designs, and ranked them...