Anthropic has released the first Claude models that add invisible watermarks to every text and file response. At the same time, the company has opened a detection API for approved organizations to verify whether content carries a Claude watermark.
Claude Fable 5.1 is now generally available through the Claude desktop app and other platforms. Claude Mythos 5.1 is limited to Anthropic’s trusted access program, which supports selected work in areas such as cybersecurity and life sciences.
The launch turns Anthropic’s earlier watermarking announcement into a product feature. The company had said that new Claude models released after its stated cutoff would include the technology. Fable 5.1 and Mythos 5.1 are the first models to do so.
Detection access for eligible organizations
The new API is available to regulators, law enforcement agencies, media organizations, fact checkers, independent researchers, educational institutions, and certain civil society groups. Companies that need to check watermarking for compliance purposes can also apply. Anthropic says it plans to expand access over time.
The API addresses an EU AI Act requirement for tools that can help authorized groups identify AI generated content. It is particularly relevant for publishers, researchers, and organizations that need to assess the origin of text.
Anthropic says its watermark does not contain information about users, organizations, or individual conversations. The system subtly influences word choices while Claude generates text, creating a pattern that the detection API can recognize. According to Anthropic, the signal can survive copying and some light editing.
A positive result indicates that Claude wrote or processed the text. A negative result, however, does not prove that a human wrote it. Heavier rewriting, translation, or other changes may affect whether a watermark remains detectable.
Sources
- Claude Fable 5.1 and Mythos 5.1 are Anthropic’s first models to watermark text outputs; a detection API is available to eligible groups as required under EU law – PCWorld
- Anthropic opens Claude AI text detection to regulators, media, fact-checkers, and others – The Decoder
Stay up to date
AI for content creation: the latest tools, tips and trends. Every two weeks in your inbox:
