Anthropic says a Claude research model designed 318,484 protein candidates in 64 hours, and 17 passed laboratory binding tests

Anthropic set an internal Claude research model the task of designing minibinders, small engineered proteins that attach to a chosen target, against five neurotoxins from scorpion species found in Mexico, Brazil, North Africa and the Middle East. Working largely without human direction, the model used 12 protein design tools to generate 318,484 candidate designs and computationally screen 553,052 protein-toxin pairs over 64 hours, Anthropic reported.

Of those, 211 designs went to the bench. Adaptyv Bio and Twist Bioscience handled synthesis and testing, and 17 met the study's binding criteria, Interesting Engineering reported. The strongest candidates bound a Mexican scorpion toxin at single-digit nanomolar affinity. Anthropic is releasing all 211 designs and the associated data, according to Runtime Wire.

Anthropic states this limitation plainly rather than omitting it: "We have not confirmed whether these binders neutralize scorpion venom in a cellular environment, so we do not claim to have produced an antivenom." Binding affinity and neutralization are separate properties, and a tight binder can attach to a toxin without blocking the part of it that causes harm.

The result is a vendor-run study rather than a peer-reviewed paper, and the 17-of-211 hit rate is the number that matters for assessing it. The scientific content is the orchestration: a model selecting and sequencing existing design tools, filtering half a million computational pairs, and choosing which 211 candidates were worth the synthesis cost. Scorpion stings affect at least 1.2 million people a year and cause more than 3,250 deaths, concentrated among children with limited access to care, Office Chai reported.