Anthropic published research on October 20 describing how an internal research model, operating within what it calls Claude Science, designed small proteins intended to bind tightly to scorpion neurotoxins. The work was done in collaboration with the investment research firm Citrini Research, with experimental testing by Adaptyv Bio and Twist Bioscience.
What Anthropic reports
Over about 64 hours, the model worked mainly autonomously with four follow-up prompts from the research team. It generated 318,484 protein designs and selected 180 for testing; the team added 31 more with the model's help. Of the 211 designs lab-tested, 17 met Anthropic's binding criteria against four scorpion neurotoxins known to be dangerous to humans. Anthropic says the model analyzed the data, generated the figures and drafted the final report under supervision.
Why scorpion toxins
Anthropic cites estimates that scorpions sting 1.2 million people or more each year, causing more than 3,250 deaths, mostly among children in regions with limited access to medical care. New treatments are expensive to produce and unlikely to earn back development costs, which is why scorpion envenomation is described as a neglected tropical disease. The project followed an earlier Anthropic experiment in which Claude designed protein minibinders from a single prompt.
What the results do and do not show
Anthropic is explicit that it has not confirmed whether the binders neutralize scorpion venom in a cellular environment, and it does not claim to have produced an antivenom. Binding to a toxin in a lab assay is an early step, not a treatment. The company says it is releasing all 211 designs and their data as a resource for the field.
The broader signal
The research is part of a wider push by model providers into scientific work, landing the same day as OpenAI's acquisition of drug-development coordination platform Orchestra Bio. For teams watching the space, the notable detail is methodological rather than medical: a general model driving a multi-day design campaign under human supervision, with outside laboratories validating the output. The distance between a designed binder and an approved treatment remains long and expensive.