Claude did not invent scorpion antivenom

Anthropic and investment research firm Citrini Research tested an unreleased version of Claude to see whether it could help design scorpion antivenom. The experiment produced protein candidates that bound to some venom toxins, but it did not create a drug or show that any candidate could neutralise venom.

Claude generated nearly 320,000 candidates over 64 hours, which were narrowed to 180 designs; the teams added 31 more. Tests by two outside laboratories found that 17 designs bound to four of the five toxins targeted, with none binding to more than one. The work cost just under $50,000 in computing and a further roughly $150,000 for testing, while experts stressed that safe, effective treatment would require much more research.