On 20 October 2026, Anthropic’s Science / life-sciences team published research on experimenting with how Claude can help accelerate drug discovery. An internal research model was asked to design minibinders—small proteins that attach tightly to a target—against scorpion neurotoxins. The initial prompt and four follow-ups were developed in collaboration with Citrini Research. Teams at Adaptyv Bio and Twist Bioscience experimentally tested 211 designs for binding; 17 met Anthropic’s binder criteria. Anthropic states clearly that it has not confirmed whether these binders neutralize scorpion venom in a cellular environment, and does not claim to have produced an antivenom—only that such binders might one day help antivenom development.
Scorpions sting 1.2 million people or more every year, resulting in more than 3,250 deaths, mostly among children in regions with limited access to care. With weak commercial incentives, scorpion envenomation has been described as a neglected tropical disease. Traditional antivenoms are typically made by immunizing animals and purifying antibodies—a months-long, costly process that usually covers only a small number of toxins and is often scarce where stings are most common.
Five toxins, four design strategies
The campaign targeted five human-dangerous scorpion-venom neurotoxins across four regions: Mexico (Cn2), Brazil (Ts1), North Africa (AaH I and AaH II), and the Middle East (LqhIII). Designs aimed at each toxin’s sodium-channel–binding site—the interface that can disrupt nerve firing and, in severe cases, contribute to heart failure and pulmonary fluid buildup.
Beyond single-toxin binders, Claude was asked to attempt multi-toxin designs: β-family (Ts1 + Cn2), α-family (AaH II / AaH I + LqhIII), and all five at once. The five toxins share one overall fold but are only about 26%–62% identical in sequence pairwise—a deliberately hard test of multi-target design.
~64 hours, more than 300,000 designs
The initial prompt defined targets, design and folding tools, and rules for novelty, scoring, and filtering. Over the next ~64 hours, an internal research model in Claude Science worked mainly autonomously, guided by four follow-ups (especially toward multi-toxin binders). Claude used 12 design tools to generate 318,484 protein designs; co-folded and screened 553,052 design–toxin pairs; and produced a final list of 180 designs. Compute was on the order of 10,000 GPU hours. The team later added 31 more designs with Claude’s help, for 211 total: 60 single-toxin (12 per toxin), 30 β-family, 30 α-family, and 91 all-five.
Dual-lab SPR: what counted as a binder
Adaptyv Bio expressed designs in a cell-free system; Twist Bioscience expressed them in mammalian cells as Fc fusions. Both measured binding by surface plasmon resonance (SPR), reporting dissociation constants (KD)—lower KD means tighter binding. To count as a binder against a toxin, a design had to be a hit at both labs, with every KD ≤ 1 µM, and show no signal against the unrelated human protein CLEC12A (LqhIII was assessed only by Adaptyv).
Figure: Left — many candidate designs converging; center — soft glowing target–binder docking metaphor; right — a few binders validated across dual labs — editorial illustration with no text or logos; not a real lab photo.
17 binders — each against a single toxin only
Of 211 designs, 17 met binder criteria, each against one toxin only: 7 against Cn2, 4 against AaH II, 4 against AaH I, and 2 against LqhIII. None of the multi-toxin designs met criteria for more than one toxin. Among the 60 single-toxin designs, 11 were binders (18%)—in line with typical protein-design campaigns, including Anthropic’s prior work.
The strongest pair, Cn2-B1 and Cn2-B2, bound Mexican scorpion toxin Cn2 with single-digit nanomolar KD at both labs. Competition experiments suggest they attach at or near the same epitope as antibody LR, which has been reported to neutralize Cn2 in mice. Anthropic stresses that Cn2-B1 and Cn2-B2 have not been tested in a cellular environment or in animals; binding to recombinant toxins is only a first step.
Outside views and open release
Outside experts Susana Vázquez Torres and Timothy P. Jenkins described the work as a promising first step: if AI can design credible binders against several hard targets in days and labs can confirm them within weeks, discovery for envenomation and other neglected diseases could, in theory, move faster. Real impact still depends on what follows binding—safe, affordable, widely accessible drugs. Anthropic is releasing all 211 designs and their binding data for others to build on.
Primary source:
- Anthropic, “How Claude designed protein binders against scorpion neurotoxins”, 2026-10-20, https://www.anthropic.com/research/claude-scorpion-toxin-binder-design
