Home / Technology / Claude 5.1 Models Debut with EU AI Act Watermarks
Claude 5.1 Models Debut with EU AI Act Watermarks
2 Sep
Summary
- Claude 5.1 models Fable and Mythos now available, with Mythos for trusted access.
- New models show double-digit performance gains in key benchmarks.
- Fable and Mythos 5.1 are the first Claude models to watermark text outputs.
Anthropic has launched Claude Fable 5.1 and Claude Mythos 5.1, with Fable 5.1 available generally and Mythos 5.1 restricted to a trusted access program. These new models represent a significant step in AI development, showcasing double-digit performance increases in agentic scientific research, agentic coding, and business workflows.
A key feature of both Fable 5.1 and Mythos 5.1 is the implementation of invisible text and file output watermarking, aligning with the requirements of the EU AI Act. This watermarking technology, mandated for new Claude models released after August 2, 2026, is undetectable to users but accessible through a dedicated detection API.
The detection API is now available to regulators, law enforcement, media, fact-checkers, and independent researchers, with plans to expand access over time. Mythos 5.1, while sharing the same core as Fable 5.1, is designed with less stringent safeguards to facilitate work in cybersecurity and life sciences, and Anthropic notes it's less prone to accessing external resources when given impossible tasks.
