Anthropic has unveiled experimental findings, demonstrating that its Claude model (in the Mythos Preview and Opus 4.8 iterations) is now proficient in autonomously executing the entire protein design workflow. During testing on 15 distinct targets, Claude achieved success in 14 cases, with 354 out of 1,320 designs confirmed as effective, culminating in an overall success rate of 26.8%. This rate significantly surpasses the industry standard of 10%-15%, with the top-performing design attaining a remarkable hit rate of 49%. Claude excelled notably in targets such as RBX1 and TNFα, with only the MBP target proving impervious to all design attempts. Furthermore, when tasked with analyzing raw data from nuclear magnetic resonance instruments, Claude Opus 5 delivered results that aligned closely with those derived from manual analysis by chemists. It is noteworthy that Claude did not innovate new tools; rather, it leveraged open-source tools, guided by approximately 16,000 words of prompts. Human involvement was primarily confined to approving network access, overseeing infrastructure, and submitting samples for testing. Although the binders designed by Claude are not yet on par with genuine drugs, Anthropic has made the relevant prompts, data, and design models publicly available, while restricting biological capabilities such as protein design in the publicly accessible version of Claude. It is remarkable to note that the journey from AlphaFold's groundbreaking protein structure predictions to Claude's autonomous protein design accomplishments spanned a mere six years.