Stock research firm used unreleased Claude to design scorpion antivenom proteins

Citrini Research sells investment notes. It has no lab, no biologists and no drug discovery experience. Yet it used an unreleased version of Anthropic's Claude to design proteins meant to grab scorpion toxins. The design work took two days. Getting real results from a lab took roughly 40 days.

What happened

The five-person team sat in a conference room at Anthropic's New York office and ran a model the public cannot use. According to Hunterbrook Media, it was supervised but lacked the bioscience guardrails that push this kind of work to a weaker public model. The first prompt ran roughly 35 pages. Claude generated 318,484 designs in 64 hours. At the team's request, Claude then emailed the sequences to Twist Bioscience.

What the lab found

Twist and Adaptyv Bio made and tested all 211 designs. Seventeen met Anthropic's binding criteria, meaning they stuck to a toxin.

• Of 60 designs aimed at a single toxin, 11 bound it, an 18% hit rate.
• None of the 151 designs aimed at several toxins bound more than one.
• Some binders were flagged as possibly sticky, so how selective they are is unconfirmed.

Anthropic does not claim to have made an antivenom. Nothing was tested for neutralizing a toxin, in cells or in animals. Critics also noted that proteins binding scorpion toxins were shown in research more than fifteen years ago.

Time and money

The lab quoted about $85,000 for the samples, which Citrini paid with an American Express card. It had roughly $200,000 committed in total. Then came the wait. Citrini said its AI biologist may have squeezed a year of work into a couple of days, but that biology works at the speed of life. The team spent about two months posting memes in Slack while it waited.

Why it matters

The only FDA-approved scorpion antivenom comes from immunizing horses. The Mexican program behind it took 25 years, then another decade of US trials. AI could one day offer a faster path. Naoual Oukkache of the Pasteur Institute of Morocco says AI-designed molecules are not about to replace today's antivenoms. For now she sees them as complementary, and they still need standardized manufacturing and clinical validation.

Susana Vázquez Torres, a researcher at CNIO, one of two outside experts Anthropic consulted, said binding is no longer the main bottleneck. The slow, costly steps are now testing and proving a design works.

What to keep in mind

This was not a story of ordinary users doing science with a public chatbot. The team used a more capable model than the public can access. Whether any of these proteins could treat a sting is still unknown.