Anthropic has published the results of an experiment in AI-driven drug discovery aimed at a neglected problem: scorpion stings, which the company says affect at least 1.2 million people a year and kill more than 3,250, mostly children in places with poor access to care. In a research post dated October 20, 2026, Anthropic reports that an internal Claude research model designed minibinders, small engineered proteins that grip a specific target, against scorpion neurotoxins, and that 17 of the designs held up in laboratory tests.
The project began when the investment research firm Citrini Research asked whether AI could let motivated non-experts tackle a real health problem. Anthropic's life-science team picked five toxins from four regions: Cn2 from Mexico, Ts1 from Brazil, AaH I and AaH II from North Africa and LqhIII from the Middle East. Claude chose the toxin structures from the Protein Data Bank and the AlphaFold database, selected binding hotspots, set its own quality thresholds and decided how to combine 12 protein-design tools.
According to Anthropic, the model ran mostly autonomously for 64 hours, steered by an initial prompt and four follow-ups, inside its Claude Science environment. It generated 318,484 candidate proteins, screened 553,052 design and toxin pairs by predicting their joint structure, and used roughly 10,000 GPU hours. Claude picked 180 designs; researchers added 31 more with its help. The contract research organisations Adaptyv Bio and Twist Bioscience then produced and tested all 211 using surface plasmon resonance.
Anthropic set a strict bar: a design counted only if both labs saw binding in every replicate, with affinity of 1 micromolar or better and no stickiness toward an unrelated human protein. Seventeen designs passed, each against a single toxin. About 18% of the single-toxin designs succeeded, which Anthropic says is typical for protein-design campaigns. None of the 91 designs meant to catch all five toxins worked against more than one.
The strongest results came against Cn2. Two binders, Cn2-B1 and Cn2-B2, showed single-digit nanomolar affinity at both labs, and competition experiments suggest they attach near the same site as LR, an antibody reported to neutralize Cn2 in mice. Antibodies and nanobodies with sub-nanomolar affinity still set the benchmark.
Anthropic is explicit about the limits. The binders were tested only against lab-grown versions of the toxins, not venom, and none has been tested on cells or in animals, so the company does not claim an antivenom. Two outside scientists, Susana Vázquez Torres of Spain's CNIO and Timothy P. Jenkins of the Technical University of Denmark, described the work as a promising first step while stressing that turning binders into safe, cheap drugs is the harder part. Anthropic is releasing all 211 designs and their data.