Claude's invisible text watermark is reaching older models, like a signature you can't see but a regulator can
By Brett Gavaghan, Founder of Adaptd ·
Anthropic watermarks text written by supported Claude models, worldwide, and it's now spreading to older models. Here's what that means for your business.
The short version
- Anthropic adds an invisible statistical watermark to text written by supported Claude models, worldwide, to meet the EU AI Act.
- The Independent reported on 3 November 2026 that it now extends to older models, including Opus 4.7, Opus 4.6 and Sonnet 4.6.
- Only eligible organisations such as regulators and researchers can detect it, so be open about AI use rather than worrying.
Text written by Claude is starting to carry a signature you can't see. It's less spy novel than it sounds, but it's worth knowing about if Claude drafts anything your clients read.
What is Claude's text watermark?
Claude's text watermark is a statistical pattern in the words Claude chooses, which detection software can use to judge whether a supported Claude model wrote a piece of text. Anthropic says nothing is added to the text, there are no hidden characters, and the watermark carries no information that traces back to a person, organisation or chat.
Anthropic added it to meet the EU AI Act's rules on labelling AI-generated content, but it applies wherever Claude is used, Australia included. Files Claude creates get a separate Content Credentials (C2PA) label.
Which Claude models does it cover?
Claude models launched on or after 2 August 2026 carry the watermark from launch, across the Claude apps, the API, Claude Code and Cowork. Anthropic is adding older models over time, and The Independent reported on 3 November that it now covers models including Opus 4.7, Opus 4.6 and Sonnet 4.6, with no way to opt out.
For now, only eligible organisations such as regulators, fact-checkers and researchers can use Anthropic's detection tool, plus companies that need it for EU compliance.
Should your business do anything about it?
Your business doesn't need to change how it uses Claude, but this is a good prompt to decide when you tell clients that AI helped. Anthropic says a watermark doesn't prove a person didn't write something, and heavily edited, paraphrased or translated text may not carry one that can be detected.
My take: treat it like a speed camera you can't see. Most of us drive the same way either way. Be open about AI use in proposals, reports and anything a client pays for, and do the checking and editing yourself.
Write your AI disclosure line now, while it's still your idea rather than somebody else's question.
Sources
Wondering what this means for your business?
That's exactly what the discovery call is for.