Anthropic announced that its new Claude Fable 5.1 and Mythos 5.1 models will embed invisible watermarks in all text and file outputs. This initiative makes Anthropic the first major AI lab to implement such transparency measures for text generation globally.
The move ensures compliance with the European Union’s AI Act, which took effect in August 2026. Anthropic will apply these watermarks to content generated for all users worldwide, rather than limiting the feature to the EU.
The statistical watermark remains imperceptible to humans and does not impact the quality or meaning of the generated text. Anthropic is releasing a private detection API to eligible groups, including regulators, law enforcement, and fact-checkers, as required by the new regulations.