Transformer News covers the results, describing binding as a first step toward antivenom treatments
Claude's autonomous scorpion neurotoxin binder campaign
Latest: Claude Autonomously Designed 17 Confirmed Scorpion Neurotoxin Binders
The campaign shows an AI system autonomously driving a protein design process from target selection through candidate generation with minimal human direction, producing designs confirmed to bind in physical lab tests. No neutralization or in-vivo efficacy tests have been conducted, so whether any binder can prevent poisoning remains unresolved.
When stories like this move, I write about them in an email each morning.
One free email each morning, around 8 AM US Eastern.
Check your email.
I have sent you a confirmation link. One click and you are in.
Not in your inbox? Look in spam or promotions. If it's there, mark it "Not spam" so future editions land in your inbox.
The full picture
Anthropic's Claude ran an autonomous protein design campaign, generating 318,484 candidate molecules targeting scorpion neurotoxins and selecting 180 for synthesis. Anthropic's team added 31 more, and all 211 were physically tested by Adaptyv Bio and Twist Bioscience. Seventeen met binder criteria. Anthropic's research page states the 17 binders each targeted one of four scorpion neurotoxins known to be dangerous to humans; one account reports five neurotoxins were targeted. Citrini Research, a finance firm with no lab or protein expertise, commissioned the work and sent only five messages to Claude throughout the entire original design process. The initial campaign concluded within 64 hours. The best-performing design, Cn2-B1, emerged during the third follow-up prompt. Anthropic is releasing all 211 designs and binding data publicly. The work is described as a demonstration of Claude's capabilities in computational protein design, not a finished antivenom.
How it developed
One of the 17 protein designs that Claude autonomously produced against scorpion neurotoxins lost 94% of its binding strength when a known venom-fighting antibody was present, suggesting the design targets the same site on the toxin as the antibody.
Anthropic published the campaign's findings on October 20, in which Claude generated 318,484 candidates and 17 of 211 physically tested designs met binder criteria.
Anthropic published findings showing Claude ran an autonomous protein design campaign against scorpion neurotoxins, generating 318,484 candidate molecules; of 211 designs physically synthesized, 17 confirmed to bind target toxins.
Citrini Research, a finance firm with no lab expertise, commissioned the work and sent only five messages over 64 hours. Testing used recombinant toxins, no neutralization tests were performed, and no binders were found for Ts1, a medically important Brazilian scorpion toxin. Anthropic is releasing all 211 designs and binding data publicly.
Sources
3 more sources
Get the daily email
When stories like this move, I write about them in an email each morning.
One free email each morning, around 8 AM US Eastern.
Check your email.
I have sent you a confirmation link. One click and you are in.
Not in your inbox? Look in spam or promotions. If it's there, mark it "Not spam" so future editions land in your inbox.