PCWorld article
Claude’s updated Fable and Mythos models watermark their replies
https://www.pcworld.com/article/3224834/claudes-updated-fable-and-mythos-models-watermark-their-replies.html
This is a news summary.
Anthropic has released its latest Claude models. As previewed last month, the new models will add an invisible watermark to all text responses in accordance with EU AI law.
Claude Fable 5.1 is currently available for general use, while Claude Mythos 5.1 is limited to participants in the company's "trusted access program". The two models are identical, but Mythos has less stringent safety measures applied to support research in cybersecurity and life sciences.
The announcement revealed various benchmarks showing performance improvements for both models, particularly double-digit score increases in the areas of agent science research, agent coding, and business workflows.
https://x.com/claudeai/status/2094848581425377479
