Claude designs proteins that bind scorpion toxins in Anthropic test
In a mostly self-directed run of about 64 hours, an internal Claude model generated 318,484 protein designs aimed at five scorpion neurotoxins from Mexico,…
October 20, 2026 · Science
In a mostly self-directed run of about 64 hours, an internal Claude model generated 318,484 protein designs aimed at five scorpion neurotoxins from Mexico, Brazil, North Africa and the Middle East. Of 211 designs tested by outside labs, 17 bound to a toxin, and the two strongest attached tightly at the same spot as a known neutralizing antibody. Anthropic stresses these are early results: the proteins have not been tested against real venom, in cells or in animals, and this is not an antivenom. All the designs and data are public.
Why it matters to you: Scorpion stings kill thousands of people a year, mostly in places where antivenom is scarce. AI that can do weeks of lab design work in days could speed up cheaper treatments.
Source: Anthropic
More from Today in AI
- New York City proposes rules to end AI customer service doom loops · October 20, 2026
- SpaceX’s AI business now earns more than Starlink, Q3 results show · October 20, 2026
- Anthropic’s Claude flags 493 security risks in open-source projects · October 20, 2026
- WhatsApp help pages hid AI instructions not to mention regulators · October 19, 2026
- EU publishes first AI Act standard, EN 18286 for quality management · October 19, 2026
All of Today in AI. Where AI is heading.
Related Power of AI pages
Keep reading with Start here, everyday uses, the tool guide, the writing workshop, and sources and standards.