What happened
Anthropic had an internal Claude research model design "minibinders" — small proteins engineered to attach to specific targets — aimed at neurotoxins in scorpion venom. Working largely autonomously with 12 protein design tools, the model generated 318,484 candidates in 64 hours, consuming roughly 10,000 GPU hours.
Lab results
A total of 211 designs, selected by Claude and supplemented by researchers, were produced and tested by Adaptyv Bio and Twist Bioscience. Seventeen met the criteria: consistent binding at both labs, passing a set affinity threshold, and no detectable binding to a control human protein. The strongest candidates bound the Mexican scorpion toxin Cn2 with single-digit nanomolar affinity.
Limits
The work is not yet a treatment. Tests used lab-produced toxins, not venom extracted from scorpions. It is not established whether the candidates block toxin activity in cells or protect animals. Clinical development would require testing venom neutralization, safety, stability and efficacy.
Open release
Anthropic released all 211 designs, predicted structures and experimental results to support further research. The findings were detailed in a technical report rather than a peer-reviewed journal.
