News · Anthropic · 2 min read
Event date: · Published:
Anthropic explains the text watermark it adds to Claude output to comply with the EU AI Act
On August 14, 2026, Anthropic published an explainer on the text watermark it adds to Claude output to comply with the EU AI Act. The company said the method needs no extra tokens, costs nothing extra and carries no identifying information.
Archive entry: this covers an event from August 14, 2026, and we published it on November 8, 2026. The details come from sources dated at the time and may have changed since.
On August 14, 2026, Anthropic published How Claude's text watermarking works, an explainer on the watermark it adds to text generated by Claude. Anthropic said it adds the watermark to comply with the EU AI Act, as several other major AI providers do.
What Anthropic said about the watermark
- Reason: as of August 2, 2026, the EU requires AI providers serving its market to mark AI-generated content. In July 2026 Anthropic signed the EU transparency code of practice for AI-generated content, with 190 other signatories.
- Output: Anthropic said the method has no practical impact on the quality or content of Claude's outputs. Nothing is added to the text, there are no hidden characters, and it needs no extra tokens and costs nothing extra.
- No identifying information: the watermark cannot be traced to a specific person, organization or chat.
- Detection: the watermark cannot be seen without Anthropic's detection API, which estimates how likely it is that Claude wrote a passage. The API was in private preview for eligible organizations.
- First models: text generated by Claude Fable 5.1 and Claude Mythos 5.1, launched on September 1, 2026, carries the watermark.
- Extension: from November 3, 2026, text generated by Claude Mythos 5, Claude Opus 4.7 and Claude Sonnet 4.6 also carries it on the platforms Anthropic lists. On Sonnet 4.6, text generated with non-default
temperature,top_portop_kmight not carry it.
What it means for people building with AI
Anyone publishing Claude-written text, including generated code comments and docs, should expect it to be machine-detectable as AI-generated. According to Anthropic, the method needs no extra tokens and costs nothing extra, so the practical difference is in detectability, not in output quality or billing.
Further Anthropic sources
Related Anthropic pages: Introducing Claude Fable 5.1 and Claude Mythos 5.1 and the Claude Platform release notes.
Sources
- How Claude's text watermarking works (Anthropic, August 14, 2026, official)
- Introducing Claude Fable 5.1 and Claude Mythos 5.1 (Anthropic, official)
- Claude Platform release notes - Claude Platform Docs (Anthropic (Claude Platform Docs), November 3, 2026, official)
Written with AI assistance, strictly from the sources listed above. Before it went live, automated checks verified its dates, sources and every language edition.