Anthropic gave investment research firm Citrini access to an unreleased Claude model to explore scorpion antivenom design. In 64 hours, the model generated nearly 320,000 candidates; outside labs tested selected proteins, and 17 bound to toxins.
That is an early research result, not a medicine. The tests did not show that the proteins neutralize toxins, and experts said safety, stability and performance in the body remain unknown. Citrini’s experiment cost about $50,000 in computing, paid by Anthropic, plus roughly $150,000 for lab testing.
Why it matters
For people following AI drug discovery, this shows Claude can help identify molecules for testing, but whether those candidates can become safe, effective antivenom remains untested.
