Citrini Research let Anthropic's Claude design scorpion antivenom: 211 proteins tested, 17 binders, then months of waiting on the lab
What happened? Citrini ResearchDealroom has a profile for this one. Try Dealroom →, the independent investment research firm best known for its AI-infrastructure work, spent three months working with Anthropic's internal research model to see whether a team of non-biologists could use AI to design a new drug. It published the results on 20 October 2026 as "Citrini Therapeutics".
How it worked: On 11 August, from a conference room at Anthropic's offices, the team prompted Claude to go from an idea to a package ready for a physical lab. Overnight, Claude spawned more than 300 sub-agents running on more than 300 concurrent GPUs to design, score and deliver a set of proteins, then sent them to a contract lab for synthesis and testing without further human input.
The results: The target was scorpion neurotoxins, which kill more than 3,000 people a year. Of the 211 designs produced and tested, 17 met the team's criteria for a binder. None bound more than one toxin, and none has yet been tested for actually neutralising a toxin, according to Nathan C. Frey, Anthropic's head of drug discovery and development research.
The catch: "Our virtual biologist may have compressed a year's worth of work into a couple days, but biology works at the speed of life," the authors write. Claude worked through problems in two days; the team then spent much of the next two months waiting on laboratory turnaround times. Citrini's takeaway is that AI in biology will progress "both alarmingly fast and not fast enough."
Why it matters: It is a rare hands-on test of AI drug design by outsiders, with real wet-lab data rather than benchmark claims, and a glimpse of Anthropic's push into life sciences. Frey joined Anthropic when it acquired Coefficient Bio, where he was CTO and co-founder.
Read more: citriniresearch.com | Nathan C. Frey on X
Image credit: Citrini Research
