Image: Anthropic

Anthropic says its Claude AI designed small proteins that grab onto deadly scorpion toxins, and lab tests backed up some of them. The company shared the work on X and in a post on its science blog. It is careful to say this is not an antivenom yet.

New on our Science Blog: designing protein binders against scorpion neurotoxins.

Scorpions sting about 1.2m people every year, and antivenoms are often costly or in short supply. Claude has made an early step toward finding new treatments.

https://t.co/l6LgdEqM7I

— Anthropic (@AnthropicAI) October 20, 2026

Scorpions sting 1.2 million people or more every year and kill more than 3,250, mostly children in places with poor access to care, according to Anthropic. Today's antivenoms come from horses. Workers milk venom from thousands of scorpions, inject it into horses, then purify antibodies from their blood. The process takes months, the products are costly, and they often run short. Drug makers have little reason to invest, because the treatment is unlikely to earn back its cost.

Anthropic tried a different route. It asked an internal research version of Claude to design minibinders. These are tiny proteins built from scratch to stick tightly to one target. A binder that covers the part of a toxin that attacks nerve cells could, in theory, stop the toxin from doing harm.

Claude worked for 64 hours with four follow-up prompts from Anthropic staff. It used 12 design tools, produced 318,484 candidate designs and picked 180 to test. Staff added 31 more with its help. The run used roughly 10,000 GPU hours of computing, a level Anthropic admits most researchers cannot afford.

Two outside labs, Adaptyv Bio and Twist Bioscience, made and tested all 211 designs. Seventeen passed Anthropic's strict test for binding, each against one toxin. The best two latch onto Cn2, a toxin from a Mexican scorpion, very tightly. Tests suggest they hit the same spot as an antibody already shown to protect mice from that toxin.

Here is the catch. No design has been tested on cells or animals, so nobody knows yet if any of them blocks venom. Claude also failed at the hardest goal. Anthropic wanted one protein that grabs several toxins at once, which would make a cheaper universal treatment. None of the 151 multi-toxin designs met that bar.

Hunterbrook, which joined the project with investment research firm Citrini, adds more caution. It reports that broader tests flagged some binders as possibly sticky, so how well they pick only their target is unconfirmed. Its CEO wrote that skilled experts could have done similar work with existing tools, though he expects Claude made it faster and cheaper. Hunterbrook also discloses that it holds trading positions related to AI in biotech.

Two outside scientists who reviewed the work called it a promising first step. Susana Vázquez Torres of Spain's CNIO cancer research center said binding is no longer the main bottleneck. The hard part, she said, is turning binders into safe, affordable drugs.

That is the real story for ordinary people. AI can now draft drug candidates for diseases that companies ignore because they do not pay. Citrini, a finance research firm whose team had no biology background according to Hunterbrook, helped steer this project. Testing, trials and approval still take years, though. Anthropic released all 211 designs and their lab data on Hugging Face so other researchers can pick up where it stopped.

For now, the advice for anyone stung by a scorpion has not changed. Get to a hospital fast. Anthropic notes that existing antivenoms work when given early.