Anthropic: Claude designs scorpion toxin binders, lab-validated

US artificial intelligence company Anthropic published a research report on October 20, 2026, stating that its internal research model Claude successfully designed protein binders (minibinders) against scorpion neurotoxins. The initiative was developed in collaboration with financial investment research firm Citrini Research, aiming to explore whether AI could enable motivated non-experts to tackle genuine problems in human health.

Scorpion stings affect more than 1.2 million people globally each year, resulting in over 3,250 deaths, primarily in regions with limited medical access. Traditional antivenoms are slow to produce, expensive, and offer low commercial returns, leading to scorpion envenomation being classified as a neglected tropical disease. The Anthropic team used Claude to analyze structures of five scorpion neurotoxins from different regions and generated 318,484 candidate protein designs.

After filtering, the team submitted 211 designs to contract research organizations Adaptyv Bio and Twist Bioscience for experimental testing. Results showed that 17 of these designs successfully bound to their target toxins. These binders targeted four known dangerous scorpion neurotoxins. Notably, this outcome was not merely a computational simulation but was validated through independent wet-lab experiments.

Despite the progress, Anthropic explicitly stated that they have not confirmed whether these binders neutralize scorpion venom in a cellular environment, and therefore do not claim to have produced an antivenom. Two external experts—Susana Vázquez Torres, a Postdoctoral Researcher in Structural Biology at the Spanish National Cancer Research Centre (CNIO), and Timothy P. Jenkins—evaluated the work as a "promising first step."