Anthropic is adding an invisible watermark to text created by its Claude models. The company confirmed the change on an updated support page this week.
The watermark applies to any model released on or after August 2. Anthropic says people will not be able to see the mark, and it will not change how the text reads.
The change is tied to new rules in Europe. The EU AI Act's Transparency Code took effect on August 2 and requires AI companies to mark content that their systems generate or edit.
How the Watermark Works
Anthropic says the watermark is built into the text itself at the model level. This means it shows up no matter which Claude product someone uses.
That includes the Claude chatbot, the API, Claude Code, Claude Cowork, and Claude Tag. The company said the mark will travel with the text if it gets copied and pasted somewhere else.
The watermark may also survive some editing, according to Anthropic. However, a heavy rewrite or a translation could remove it.
It is still unclear exactly how much editing is needed to strip the mark out. TechCrunch reported it had asked Anthropic for more detail and was waiting on a response.
Anthropic said it plans to extend the watermarking system to older Claude models over time. Right now, the feature only applies to models released after August 2.
Images are covered too. Anthropic said image files processed by Claude will carry a watermark using the C2PA standard, an open system used to track digital content.
That image watermark shows that Claude touched the file in some way. It can also flag if the file has been changed since Claude processed it.
Other Companies Making Similar Moves
Anthropic is not alone in rolling out this kind of tool. Other companies have promised to follow the EU's transparency code as well.
Those companies include Black Forest Labs, Google, Meta, Microsoft, OpenAI, and Synthesia. Each has agreed to mark AI-generated content in some way.
Anthropic pointed out that its watermark only shows that Claude played some role in a piece of text. It does not mean Claude wrote the entire thing from scratch.
Even a small task, like asking Claude to proofread or translate a paragraph, could leave a trace behind. That applies across all the surfaces where people use Claude.
Other platforms have made similar changes recently. Music platform Suno said last week it would begin marking AI-generated songs after facing legal challenges.
Newsletter platform Substack also teamed up with a company called Pangram last month. That partnership lets Substack flag newsletters that were written with AI help.
Anthropic's watermark rollout is part of a broader wave of companies adding these tools in 2026. The changes are being driven largely by new European regulation that took effect this month.
