Anthropic reported on October 20, 2026, that 17 of 211 Claude-designed candidates met its criteria for binding scorpion neurotoxins in laboratory tests. That is about 8.1% of the tested designs. The result concerns binding, not a demonstrated antivenom: the candidates were not tested for toxin neutralization in cells or efficacy in animals.
From computer-designed proteins to lab tests
Anthropic says its internal Claude research model generated 318,484 small proteins, called minibinders, over 64 hours. It used 12 design tools and screened 553,052 design-toxin pairs, with about 10,000 GPU-hours of computation. Claude selected 180 designs for testing, and Anthropic’s team added 31 with Claude’s help.
Adaptyv Bio and Twist Bioscience tested the resulting 211 designs. The computational work produced candidates; the partner laboratories then tested whether they bound the target toxins.
How many designs met the criteria for each toxin?
Anthropic reported 17 qualifying designs across five targets. The counts add up to 17, and each qualifying design met the criteria against one toxin only.
| Scorpion neurotoxin | Designs meeting Anthropic’s binding criteria |
| Cn2 | 7 |
| Ts1 | 0 |
| AaH I | 4 |
| AaH II | 4 |
| LqhIII | 2 |
The 17 qualifying designs represent about 8.1% of all 211 tested. Anthropic separately reported that 11 of the 60 designs aimed at a single toxin qualified, or 18%, and said that result was consistent with typical protein-design campaigns. The 18% figure applies to that subset, not the full set of tested designs.
Anthropic says none of the designs aimed at multiple toxins qualified against more than one. Adaptyv Bio alone assessed LqhIII.
What counted as a binder?
Anthropic’s criteria required evidence of binding in every replicate at both laboratories, a measured KD of 1 micromolar or tighter for each result, and no signal against CLEC12A, an off-target protein. KD is a measure of binding affinity; a lower value means tighter binding. Anthropic says Adaptyv Bio used cell-free expression, while Twist Bioscience expressed the designs in mammalian cells as antibody-Fc fusions. Both laboratories used surface plasmon resonance (SPR), a method for measuring binding between molecules.
The assays used recombinant toxins. Anthropic reports that LqhIII was assessed only by Adaptyv Bio, rather than at both laboratories.
Binding does not establish toxin neutralization
The designs targeted the site where the toxins bind to sodium channels, which carry electrical signals in nerves. A binder attaching at or near that site could be relevant to blocking the toxin’s interaction, but the reported tests measured binding. Anthropic says it did not test whether the candidates neutralized toxin in cells or protected animals, and it does not claim to have produced an antivenom.
Anthropic reported single-digit nanomolar binding for Cn2-B1 and Cn2-B2 in tests at both partner laboratories. Competition experiments suggested that the candidates bind on or near the site used by antibody LR. For Cn2-B1, Anthropic described testing against toxin purified from venom, patch-clamp recordings of sodium-channel activity, and mouse studies as next steps. It also said it released the 211 designs and their binding data.
