Anthropic: Claude designed protein binders against scorpion neurotoxins, 17 of 211 passed lab tests

Anthropic published a research report on 20 October 2026 describing how an internal Claude research model designed small proteins (“minibinders”) meant to latch onto scorpion neurotoxins — a possible first step toward cheaper antivenoms. The project was developed with Citrini Research, and lab testing was done by Adaptyv Bio and Twist Bioscience.

What was done

• Claude targeted five neurotoxins from scorpions in Mexico (Cn2), Brazil (Ts1), North Africa (AaH I, AaH II) and the Middle East (LqhIII).
• Working mainly autonomously over 64 hours, with four follow-up prompts, it used 12 design tools to generate 318,484 designs and chose 180 for testing; the team added 31 more with Claude’s help (211 total). Anthropic says the campaign used on the order of 10,000 GPU hours.
• 17 designs met strict binding criteria, each against a single toxin. None of the designs aimed at several toxins at once worked against more than one.
• The two strongest binders, against the Mexican toxin Cn2, bound with affinities in the single-digit nanomolar range and appear to compete for the same site as an antibody reported to neutralise Cn2 in mice.

Why it matters

Anthropic cites estimates that scorpions sting 1.2 million or more people a year, with more than 3,250 deaths, mostly children in areas with poor access to care. All 211 designs and data are being released. Outside expert Susana Vázquez Torres (CNIO) said binding “is no longer the main bottleneck” — the hard part is turning binders into safe, affordable drugs.

Limitations

This is a company report, not a peer-reviewed paper. Binding was measured only against lab-made (recombinant) toxins; the designs were not tested for neutralisation in cells or for protection in animals, and Anthropic explicitly says it does not claim to have produced an antivenom.