Claude Help Center article How Claude marks AI-generated content
Claude Help Center on marking AI-generated content, captured 2026-11-02. Image: AppFor.ai screenshot of Anthropic's page. Image source ↗

What happened

Anthropic announced text watermarking in a post dated August 14, 2026 and updated it on September 1 with details on the detection API. It has signed the EU AI Act's Article 50(2) Code of Practice on Transparency of AI-Generated Content. Its Help Center now explains that Claude models launched in the EU on or after August 2, 2026 support machine-readable marking at launch: generated text carries embedded watermarks, and generated files include Content Credentials (C2PA) where supported.

Marks apply across Claude, the Claude Platform API, Claude Code, Claude Cowork and Claude Tag, worldwide, and on AWS, Google Cloud and Microsoft Foundry for supported models. Models released before August 2, 2026 fall under a transition period. The Help Center table now lists text watermarks for older models including Opus 4.8, Sonnet 4.6 and 4.5, Haiku 4.5 and Mythos 5, alongside current models such as Opus 5.5, Sonnet 5.5 and Fable 5.1.

How the watermark works

Anthropic says nothing is added to the text and there are no hidden characters. The technique, introduced in Google DeepMind's SynthID-Text paper, nudges Claude's word choices where several are equally good. It needs no extra tokens, so it does not change price or speed, and it carries no information that could trace text to a person, organization or chat.

Because the mark lives in word choices, it is sparser in factual passages and code, where there is often one right answer. When Claude only proofreads human writing, there may be too little of Claude's own wording for the mark to be detectable.

What it means for you

For most users nothing changes in daily use. If you publish text Claude wrote, assume that authorized parties may be able to detect it. Detection is in private preview for organizations entitled under EU law, such as regulators, media, fact-checkers and researchers, and for companies that must verify marking for compliance.

Anthropic has not published a public detector, and says it will update its guidance as more technical details become available.

Sources & further reading

Based on the official documentation below. Features and access can change; check the provider before making a decision.

  1. Claude Help Center: How Claude marks AI-generated content (opens in a new tab)
  2. Anthropic: Claude text watermark (opens in a new tab)
Something out of date or incorrect?Report a correction

Was this page useful?

Help us make our coverage more useful. This is feedback on the page, not a rating of the tool.

Share this page

Related reading