Anthropic has shown that its internal Claude research model can largely autonomously design protein binders against scorpion neurotoxins, an AI-for-science result that points toward faster, cheaper drug discovery. Working inside Claude Science, the company's AI environment for researchers, the model generated 318,484 candidate designs and selected 180 for laboratory testing, according to Anthropic's technical report published October 20, 2026.
The project, run with the financial research firm Citrini Research, targeted five neurotoxins from four scorpion species. Two contract research organizations, Adaptyv Bio and Twist Bioscience, tested 211 designs, and 17 met Anthropic's criteria for a binder. The company says it has not confirmed whether these binders neutralize venom in cells, so it does not claim to have produced an antivenom.
How Claude designed the binders
Anthropic gave Claude the same prompt used in an earlier protein-binder study, modified with Citrini. Over 64 hours, the model worked mainly on its own, directed by four follow-up prompts. It analyzed data, generated figures, and drafted the final report under human supervision, Anthropic says.
The goal was minibinders, small proteins that attach tightly to a target. Scorpion neurotoxins that harm humans bind to sodium channels in nerve cells, disrupting signals and, in severe cases, causing heart failure or fluid buildup in the lungs. A binder that blocks the toxin's docking site could, in theory, stop that chain.
Claude scorpion antivenom results
Of 211 lab-tested designs, 17 met the criteria, and the two strongest bound extremely tightly to Cn2, a toxin from a Mexican scorpion, according to The Decoder. Lab tests suggest they dock at or near the same site as an antibody known to neutralize Cn2 in mice. Two independent labs confirmed the results, the report says.
