Anthropic: an internal Claude model designed protein binders against scorpion neurotoxins

Anthropic published a research post on October 20, 2026. Everything below is Anthropic's own account.

• Setup: an internal research model, working in Claude Science, was prompted (plus four follow-ups) to design minibinders against five scorpion-venom neurotoxins (Cn2, Ts1, AaH I, AaH II, LqhIII). The work was done with Citrini Research. Anthropic says the campaign ran about 64 hours and used on the order of 10,000 GPU hours, with 12 design tools and 318,484 designs generated.
• Lab test: 211 designs were tested for binding by Adaptyv Bio and Twist Bioscience; 17 met Anthropic's criteria, each against a single toxin (7 for Cn2, 4 each for AaH II and AaH I, 2 for LqhIII). None of the designs aimed at several toxins met the criteria for more than one. The two strongest, against Cn2, bound with single-digit nanomolar affinity.
• Limits stated: binding was measured only against recombinant toxins. Neutralization in cells or animals was not tested, and Anthropic says it does not claim to have produced an antivenom. Two outside experts called it a promising first step.
• Anthropic is releasing all 211 designs and their data, plus a technical report.