Claude designed 211 proteins to block scorpion venom. 17 worked in the lab
In 64 hours, mostly on its own, an internal Claude model generated 318,484 protein designs aimed at scorpion neurotoxins. Lab tests confirmed 17 binders. Anthropic is clear it has not made an antivenom.
By ScoviAI Team
Key points
- Published by Anthropic on October 20, 2026
- An internal Claude research model worked mostly autonomously for 64 hours, with four follow-up prompts
- 211 designs were lab-tested by Adaptyv Bio and Twist Bioscience; 17 met the criteria for a binder
- All 211 designs and their data are being released; no one has shown they neutralize venom yet
Scorpions sting 1.2 million people or more every year, and more than 3,250 of them die, mostly children in places with little access to medical care. New treatments are expensive to develop and unlikely to pay for themselves, which is why scorpion stings are sometimes described as a neglected tropical disease.
Anthropic wanted to see whether its AI could make a dent in that problem. The result, published October 20, is 17 lab-confirmed protein binders against scorpion neurotoxins, designed largely by Claude.
How the experiment worked
The question came from an unexpected place. A team from Citrini Research, a financial investment research firm, asked Anthropic whether AI could let motivated non-experts tackle a real health problem. Anthropic's life sciences team picked scorpion venom with Claude's help.
An internal research model, running inside Anthropic's Claude Science environment, then worked mainly on its own for 64 hours. People steered it with four follow-up prompts. Along the way it:
- searched the Protein Data Bank and the AlphaFold database for toxin structures,
- generated 318,484 protein designs,
- picked 180 for testing (the team added 31 more, with Claude's help),
- analyzed the results, made the figures and drafted the report.
Two contract research labs, Adaptyv Bio and Twist Bioscience, tested all 211 designs. Seventeen met Anthropic's criteria for a binder, each attaching to one of four neurotoxins known to be dangerous to people.
Which toxins were targeted?
The team aimed at five toxins from scorpions in four regions:
| Region | Toxin |
|---|---|
| Mexico | Cn2 |
| Brazil | Ts1 |
| North Africa | AaH I and AaH II |
| Middle East | LqhIII |
These toxins latch onto sodium channels in nerve cells and make nerves fire at the wrong time. In severe cases that can lead to heart failure and fluid in the lungs. The designs target the exact spot each toxin uses to grab the channel.
ADVERTISEMENT
What it doesn't prove
Anthropic is careful here. It has not confirmed that the binders neutralize venom in cells, let alone in animals or people, "so we do not claim to have produced an antivenom." Today's antivenoms are made by immunizing horses with venom milked from thousands of scorpions, take months to produce and often cover only a few toxins.
Two outside scientists reviewed the work. Susana Vázquez Torres of the Spanish National Cancer Research Centre said that getting very tight binders is no longer the main bottleneck, and that the real challenge is turning them into "safe, affordable, and widely accessible" drugs.
Why it matters
If an AI system can design credible binders against hard targets in days, and labs can check them in weeks, early drug discovery for neglected diseases could speed up considerably. Anthropic is releasing all 211 designs and their data so other researchers can build on them. It's part of a wider push into science; for a very different example, see how Hugging Face's Carbon-X reads DNA.
Sources
ScoviAI Team is the ScoviAI newsroom. We write every article ourselves from the original announcements and link our sources. Spotted a mistake? Tell us and we'll fix it.
ADVERTISEMENT
