Claude designs validated protein binders against scorpion neurotoxins
Anthropic reported on 20 October 2026 that an internal Claude research model, running in its Claude Science environment with Citrini Research, autonomously designed protein minibinders against five neurotoxins from four scorpion species. Over about 64 hours, prompted once at the start and four times afterwards, the model generated 318,484 designs across more than 550,000 design-toxin pairs, chose the toxin epitopes to target, ran twelve design programmes, set its own scoring cutoffs and selected 180 candidates for synthesis; Anthropic's life-science team added 31 more. Of the 211 designs tested by Adaptyv Bio and Twist Bioscience, 17 met the binding criteria, with the strongest binding a Mexican scorpion toxin at single-digit nanomolar affinity. Eleven of 60 single-toxin designs bound their target; no design intended to bind multiple toxins bound more than one. Anthropic released all 211 designs and the laboratory data publicly.
Why It Mattered
The result is modest as medicine and significant as a demonstration of autonomy. Binding is the first of many steps toward an antivenom: there is no neutralisation assay, no animal work and no evidence that any of the seventeen molecules would help a person who has been stung. What changed is the degree of unsupervised scientific judgement exercised between the prompt and the wet lab. Earlier AI protein design, including the work recognised by the 2024 chemistry Nobel, placed models inside pipelines whose targets, scoring functions and candidate selection were set by humans. Here the model made those choices, and an external laboratory returned an 18 percent hit rate on single-toxin designs that is comparable to conventional campaigns run by specialists. Scorpion envenomation is a neglected condition — roughly 1.2 million stings a year, concentrated in regions where antivenom is scarce or unaffordable — which is precisely the class of problem that attracts little industrial effort. If an agentic system can compress the design stage of such campaigns to days at trivial cost, the economic argument for ignoring neglected targets weakens. The same demonstration carries the obvious dual-use reading, and it will be cited in both directions. A system that can design high-affinity binders to venom proteins with minimal human direction is exercising capabilities adjacent to those that biosecurity frameworks are written to constrain, and the publication of the full design set is itself a disclosure decision. Anthropic framed the work as an early step rather than a treatment, and the careful hedging in the announcement is part of why the episode is likely to be referenced: it is a reasonably well-documented baseline for what autonomous molecular design could do in late 2026.
Who Built It
Anthropic (Claude Science), with Citrini Research, Adaptyv Bio and Twist Bioscience
Applications
- Protein Design
- Drug Discovery
- Neglected Tropical Diseases