An Anthropic Model Designs Proteins That Stick to Scorpion Toxins

In 64 hours, an Anthropic research model designed small proteins meant to stick to scorpion toxins. Two labs tested 211 designs and 17 met the binding bar, but none has been tested for stopping a toxin in cells or animals.
artificial-intelligence
data-science
Author

Kabui, Charles

Published

2026-11-02

Keywords

scorpion-toxins, protein-design, de-novo-binders, antivenom, claude-science, ai-for-drug-discovery

Download as PDF

Anthropic used an internal AI research model to design small proteins from scratch (“de novo”) that stick tightly to scorpion toxins, a step toward new antivenoms. Inside Claude Science, its workbench for scientists, the model spent 64 hours generating 318,484 designs and picking 180. The team added 31 more. Two contract labs, Adaptyv Bio and Twist Bioscience, tested all 211 against their five target toxins from four regions. A binder had to hold on in every repeat at each lab that tested it, with a strength of 1 micromolar or better (a lower number is a tighter grip), and show no grip on an unrelated human protein. Seventeen passed.

A 2008 review estimates over 1.2 million scorpion stings a year and more than 3,250 deaths, with children hit hardest. Antivenoms are typically made from horses injected with venom milked from thousands of scorpions. That can take months, and each covers only a few toxins. Outside experts Susana Vázquez Torres and Timothy Jenkins called it a promising first step. Anthropic released all 211 designs and the lab data.

No binder has been tested in cells or animals for neutralizing a toxin, meaning stopping it from reaching nerve cells. Every test used lab-made toxins, not venom, so Anthropic does not claim an antivenom. The two best, both against the Mexican toxin Cn2, gripped more than 100 times tighter than the 1 micromolar bar. None of the 151 designs aimed at several toxins qualified as a binder for more than one.

Read More: Claude Science builds a complete ultraviolet sky map and OpenAI’s GPT-Rosalind, a model for biology and drug discovery

Sources:

Disclaimer: For information only. Accuracy or completeness not guaranteed. Illegal use prohibited. Not professional advice or solicitation. Read more: /terms-of-service

Reuse

GNU GENERAL PUBLIC LICENSE v3.0(View License)

Citation

BibTeX citation:
@misc{kabui2026,
  author = {{Kabui, Charles}},
  title = {An {Anthropic} {Model} {Designs} {Proteins} {That} {Stick} to
    {Scorpion} {Toxins}},
  date = {2026-11-02},
  url = {https://toknow.ai/posts/claude-scorpion-neurotoxin-binders-211-designs/},
  langid = {en-GB}
}
For attribution, please cite this work as:
Kabui, Charles. 2026. “An Anthropic Model Designs Proteins That Stick to Scorpion Toxins.” https://toknow.ai/posts/claude-scorpion-neurotoxin-binders-211-designs/.