Anthropic published a research result on Monday that grew out of an unusual question: could motivated nonexperts use Claude to tackle a real problem in human health? Citrini Research, an investment research firm, asked it. The answer came back as 17 protein binders against scorpion neurotoxins.
What the model actually did
An internal research model worked for 64 hours inside Claude Science, mostly without intervention. It got a starting prompt—adapted from an earlier binder study together with Citrini—and four follow-up messages. That was the whole steering effort.
In that window, Claude generated 318,484 protein designs and picked 180 of them for lab testing. The team added 31 more, also with Claude’s help. Contract research organizations Adaptyv Bio and Twist Bioscience tested all 211 designs to see whether they bound to their target toxin. Seventeen met the criteria, spread across four neurotoxins known to be dangerous to humans. Claude also ran the data analysis, produced the figures and drafted the final report, under supervision.
Not an antivenom, and Anthropic says so
Whether those 17 binders actually neutralize venom inside a cell has not been tested. Anthropic explicitly makes no claim to have produced an antivenom. All 211 designs and their measurement data are published.
Why scorpions? More than 1.2 million people are stung every year and over 3,250 die, mostly children in regions with poor access to care. Conventional antivenoms are made by immunizing horses and purifying their antibodies: months of work, expensive, and each serum covers only a handful of toxins. The economics never add up, which is why envenomation counts as a neglected tropical disease.
Two outside experts reviewed the work, Susana Vázquez Torres of Spain’s national cancer research centre CNIO and Timothy P. Jenkins. Both call it a promising first step.
Binding is no longer the hard part
Vázquez Torres says the line that matters: designing a binder against almost any target no longer seems unrealistic, and binding itself is no longer the main bottleneck. That is exactly where the limit sits. Everything between a protein that binds and a drug that works—efficacy in a body, safety, approval, a price a clinic in North Africa or India can pay—Claude never touched. That this part is now the hardest one is still good news. Two years ago the bottleneck was the design.