Five people at a stock research firm used Claude to design scorpion antivenom candidates, and their first request hit a safety wall

Citrini Research, an investment research firm with almost no scientists on staff, used an internal Anthropic research model to design proteins that bind scorpion toxins, and got them tested in a lab.

Anthropic said on its science blog on Oct 20 that Claude produced 318,484 protein designs over 64 hours and picked 180; with 31 more added by the team, 211 were lab-tested and 17 met the bar for a binder. They bound toxins from scorpions in Mexico, North Africa and the Middle East, but not a Brazilian species. Anthropic released all 211 designs and the lab data.

Per Citrini's write-up, three of the five team members had no science background at all. They budgeted $200,000, and their first prompt was immediately blocked by biosecurity guardrails, so they went to Anthropic's offices and ran the work on its internal network and machines. The prompt, drafted with Claude, ran about 35 pages; compute cost $49,997 and the lab work just under $150,000.

It is not an antivenom yet. Anthropic said it is "not claiming to have produced an antivenom," with cell and animal tests still to come. One X reply noted that "binding is the cheap half of an antivenom"; Citrini answered that waiting for functional assays would have meant another two to three months. Citrini's announcement passed 720,000 views.

Scorpions sting more than 1.2 million people a year, killing over 3,250. Today's antivenoms are made from antibodies harvested from horses injected with venom.