Anthropic announced on October 20, 2026, that 17 of 211 laboratory-tested, Claude-designed proteins met its criteria for binding scorpion toxins. The tests measured binding to recombinant toxins, not whether the proteins neutralize venom or protect an animal.
What counted as a qualifying binder
Anthropic counted a design when tests found binding in every replicate at the testing organizations, each measured binding affinity (Kᴅ) was 1 micromolar or tighter, and the candidate showed no signal against CLEC12A, an unrelated human protein. A lower Kᴅ indicates tighter binding. Anthropic noted that LqhIII was assessed only by Adaptyv Bio.
The tests used surface plasmon resonance, a method for measuring how strongly molecules bind. They used recombinant toxins rather than toxins purified from scorpion venom. Each of the 17 designs met the criteria for one toxin; none qualified against more than one.
The 17 qualifying designs targeted four toxins
| Toxin target | Designs meeting Anthropic’s criteria |
| Cn2 | 7 |
| Ts1 | 0 |
| AaH II | 4 |
| AaH I | 4 |
| LqhIII | 2 |
Anthropic said Adaptyv Bio alone assessed LqhIII. The other target counts are Cn2, associated with Mexico; Ts1, associated with Brazil; AaH I and AaH II, associated with North Africa; and LqhIII, associated with the Middle East.
Claude designed candidates; laboratories tested them
Anthropic reported that its internal research model generated 318,484 protein designs over a 64-hour campaign, using 12 design tools and computationally screening 553,052 design-toxin pairs. Claude selected 180 designs for testing; Anthropic’s team added 31, bringing the laboratory set to 211. The campaign date was not specified.
Citrini Research helped shape the prompt. Anthropic said researchers supervised the work and gave Claude four follow-up prompts. The model handled the computational design and screening; Adaptyv Bio and Twist Bioscience performed the physical laboratory work.
Adaptyv Bio used a cell-free bacterial system to make proteins. Twist Bioscience expressed the designs in human kidney cells as fusions with an antibody’s Fc region. Both organizations measured binding with surface plasmon resonance.
Binding is an early research result, not an antivenom
Anthropic reported single-digit nanomolar binding affinity for Cn2-B1 and Cn2-B2 in both organizations’ assays. Competition experiments suggested that these designs bind at or near the site targeted by antibody LR, which has been reported to neutralize Cn2 in mice. The experiments did not test whether Cn2-B1 or Cn2-B2 neutralizes toxin.
Anthropic did not test the new designs for toxin neutralization in cells or protection in live animals, and said the work had not produced an antivenom. It outlined further research: test binding to toxin purified from venom, measure effects on sodium channels in cells using patch-clamp recordings, and then test protection in mice. Safety, stability, affordability and access would also matter in developing a treatment.
Anthropic made all 211 designs and their binding data available in its public dataset.
