Citrini Research used Anthropic’s internal model to design novel proteins for a scorpion antivenom. The team had no biology background. They ran the project on Anthropic’s local network to meet biosecurity rules. Claude spawned over 300 sub-agents across 300 GPUs to design and score candidates overnight. The proteins were sent to a lab for synthesis and testing without human input. The goal was a minibinder targeting scorpion toxins. They also attempted a single molecule to bind five toxins as a moonshot. The project cost $200,000 and aimed to test AI’s role in drug discovery. Scorpion venom is the most valuable liquid in the world at $39M per gallon. The team concluded AI in biology will progress both fast and not fast enough.
What changed
AI drug design required deep domain expertise and manual intervention at every step.
What it unlocks
A team with no biology background used AI to design, test, and validate novel proteins autonomously.
- $200,000 budget
- 300+ sub-agents
- 300+ concurrent GPUs
- $39M per gallon of scorpion venom
What you need to act on it
- access to an internal research model
- on-site execution at Anthropic offices
- biosecurity questionnaire compliance
Sources