Claude Designs 17 Scorpion Venom Protein Binders in 64 Hours
Anthropic's Claude AI generated 318,484 protein designs and 17 were confirmed to bind scorpion neurotoxins after independent lab testing.
Anthropic's Claude AI generated 17 protein designs confirmed to bind dangerous scorpion neurotoxins in a 64-hour autonomous design campaign. The results, published October 20, 2026, were verified by two independent labs, Adaptyv Bio and Twist Bioscience.
1. What happened
Anthropic asked an internal Claude research model to design small proteins called "minibinders." A minibinder is a protein that attaches tightly to a specific target, in this case the part of scorpion venom that attacks nerve cells.
Claude worked almost entirely on its own. It received one initial prompt and four follow-ups, then generated 318,484 protein designs using roughly 10,000 GPU hours. From that total, Claude selected 180 designs for testing. The Anthropic life sciences team added 31 more with Claude's help, bringing the total to 211 designs sent to labs.
Adaptyv Bio and Twist Bioscience, two contract research organizations, tested all 211 designs. Seventeen were confirmed to bind the toxins according to the set criteria.
2. Why scorpions
Scorpions sting more than 1.2 million people each year and cause more than 3,250 deaths. Most victims are children in regions with limited medical access.
Existing antivenoms are expensive and slow to produce. The traditional process requires venom milked from thousands of scorpions, horses immunized with that venom, and antibodies purified from horse blood. This takes months and only works for a narrow set of neurotoxins.
Anthropic and their partner, investment research firm Citrini Research, chose scorpions because envenomation is described as a neglected tropical disease. Commercial investment in scorpion antivenom is very low, so an AI approach could open a new path.
3. Five toxins, four regions
Claude targeted five neurotoxins from four regions: Cn2 from Mexico, Ts1 from Brazil, AaH I and AaH II from North Africa, and LqhIII from the Middle East.
The breakdown reported by TokenPost: seven designs bound Cn2, four bound AaH I, four bound AaH II, and two bound LqhIII. No design qualified for Ts1 from Brazil.
The two strongest binders, called Cn2-B1 and Cn2-B2, showed binding affinities in the single-digit nanomolar range, a measure of how tightly a protein attaches to its target.
Claude also attempted to design proteins that bind multiple toxins at once, but none succeeded. Each of the 17 confirmed proteins bound only a single toxin.
4. Not an antivenom yet
Anthropic stated clearly that they have not produced an antivenom. The tests only measured whether the proteins attach to their targets, not whether they stop the toxin from working.
The toxins tested were produced in the lab from toxin genes, not purified venom. The strongest binders have not been tested in living cells or animals.
The full pipeline to a real drug remains long. A binder would need to be confirmed against venom-purified toxin, tested in living cells using "patch-clamp" recording to see if it blocks the toxin from attacking sodium channels, then tested in mice. After that come safety, stability, and manufacturing studies.
5. Data released openly
Anthropic released all 211 designs and their binding data for free on Hugging Face. Other research teams can use this data to continue the work.
Two outside experts, Susana Vazquez Torres from the Spanish National Cancer Research Centre and Timothy P. Jenkins from the Technical University of Denmark, commented that the work shows AI can design credible protein binders against multiple toxins in days, a process that previously took months.
Citrini Research, which started the project with a USD 200,000 budget, noted that their team had no science background. They wanted to test whether AI could genuinely let nonexperts tackle a real health problem.
What it means for you
AI can now generate lab-confirmed protein designs in days rather than months. This accelerates the earliest stage of drug discovery for diseases that have long been ignored because they are not commercially profitable.
- Faster drug research: Credible binders can be created in 64 hours, an early step that previously took months. This does not mean a drug is ready, but the starting point is far quicker.
- Neglected diseases get attention: Scorpions kill 3,250 people a year, mostly children. AI can shift the cost equation for diseases that pharmaceutical companies will not touch.
- Nonexperts can try: Citrini Research, an investment firm with no biology background, produced lab-confirmed proteins. This lowers the barrier to entry for scientific research.
- Open data for everyone: All 211 designs are released free. Other teams can continue from here without repeating the early work.
- Stay cautious: A binder is only the first step. Testing in living cells, animals, and human clinical trials is still needed before any real drug exists. Anthropic itself does not claim to have created an antivenom.
For AI tools you can use daily, visit AI Tools.