Claude fails to design scorpion antivenom
In brief
- Citrini Research tested an unreleased version of Anthropic’s Claude to design scorpion antivenom.
- The AI generated 320,000 protein candidates, but none led to a viable drug, experts say.
- Lab-tested proteins neutralized only one toxin at a time, far from a functional treatment.
A financial research firm tried to prove AI could revolutionize drug discovery by testing an unreleased version of Anthropic’s Claude on scorpion antivenom design. According to *The Verge*, the experiment fell short: experts say the AI delivered promising leads, but none viable for an actual drug.
Citrini Research’s trial used an advanced Claude model for 64 hours, generating nearly 320,000 protein candidates. After filtering out toxic or known structures, 211 designs remained. These were tested by two labs, Adaptyv Bio and Twist Biosciences, to check their ability to neutralize venom. Sixteen proteins showed affinity with four of five targeted toxins, but none bound to multiple toxins at once.
Experts stress that designing binding proteins is the “easy” part compared to creating a functional drug. Despite the failure, Citrini frames the experiment as a success, highlighting investment opportunities in AI-driven drug discovery. Anthropic did not comment on the trial.
What it changes for you
- HealthThe experiment highlights AI's potential to speed up the discovery of binding proteins, but translating these into effective drugs remains a major challenge for the sector.
This summary gives you the essentials. The full article, its figures and quotes are on AI | The Verge.
Read the original articleAI-generated summary of the source, checked by rules. How we write · Report an error