Claude Protein Binders: Anthropic has claimed that two of its frontier Claude models (Mythos Preview and Opus 4.8) have designed de novo protein binders, an early foundation step in clinical drug making, and have taken lesser time and delivered better results than typical protein design campaigns.
In a post on Wednesday, Anthropic shared that Claude autonomously designed protein binders from scratch (de novo) against 14 out of 15 targets, with 22-35% of its individual designs binding successfully. Protein binders are protein-like molecules specifically engineered or selected to attach tightly to a particular target molecule. Its considered as the first step in the process of clinical drug making.
The results by Claude were then independently synthesized and tested by Adaptyv Bio and Twist Bioscience labs where Claude had an overall hit rate of approximately 27% (354 confirmed binders). Multi-target campaigns (all targets simultaneously) yielded 22.6–26.7% success, while single-target mode reached 35.1%.
Anthropic informed that Claude received a detailed human-written prompt and then operated largely autonomously by researching targets, selecting binding sites (epitopes), orchestrating open-source structure and sequencing design tools, optimizing candidates in silico, and generating ranked designs while screening for solubility, expressibility, and novelty.
Many drugs work by binding to a specific target in the body and blocking or changing what it does. An important first step in the drug development process is designing a molecule that can bind tightly to its target. Traditionally, that’s meant weeks or months of expert work per… pic.twitter.com/CGCNTNaKBq
— Anthropic (@AnthropicAI) August 18, 2026
“Our external evaluators, Adaptyv Bio and Twist Bioscience, independently produced and tested Claude’s designs in the lab, finding that of the 15 targets we designed against, Claude successfully designed binders against 14 of them. These include high-affinity binders1 against at least six targets, and binders matching or exceeding the best reported affinity against at least four targets..Mythos Preview and Opus 4.8 achieve overall hit rates—how many of the designs are, in fact, binders—of 26.7% and 22.6%, respectively, when designing against all targets simultaneously in a 48-hour session. 10 to 15% is typical in protein design campaigns today,” read a statement from Anthropic.
The company informed that AI models struggled against BBF-14, a β-barrel-shaped protein that was itself de novo designed and maltose-binding protein–a large flexible bacterial protein.
Anthropic benchmarked their newest Claude models on protein engineering tasks, and we at @adaptyvbio ran the wet lab work behind it.
They picked 16 targets from our past protein design competitions on @proteinbase and sent us an anonymised list of designs, so we had no idea… https://t.co/WMxJnA3hPp pic.twitter.com/BBiRPHTffK
— Julian Englert (@julian_englert) August 18, 2026
Anthropic also claimed that they are now training Claude to run the entire drug development process end-to end.
“As we work to deliver these capabilities safely via trusted access programs, protein design and other dual-use research biology capabilities remain unavailable for general access in Claude Fable 5,” informed Anthropic.
Also Read: AI Wrote 16 Working Viruses. The Rules Haven’t Caught Up
