Anthropic Introduces Invisible Watermarks To Identify AI Content

Anthropic, in an official blog post has announced that Claude can embed invisible watermarks in AI-generated text, creating a hidden trail that can survive copy-pasting.

So, if you ask Claude to write something for you, there could now be a hidden marker sitting inside that text.

You won't see it. It won't change how the text reads. But Anthropic said supported Claude models can embed an imperceptible, machine-readable watermark directly into AI-generated text.

And because the watermark is part of the text itself, it can travel with the text when it is copied and pasted elsewhere, and may survive some editing.

The move is part of Anthropic's commitments under the EU AI Act's Code of Practice on Transparency of AI-Generated Content. Claude models launched in the European Union on or after August 2, 2026 will support machine-readable marking from launch. Anthropic says the marking will also apply worldwide to supported models.

The system covers Claude's API as well as products including Claude, Claude Code, Claude Cowork and Claude Tag.

But text isn't the only thing Claude can mark. For supported files such as PNG, JPG and SVG, Anthropic said Claude will attach digitally signed provenance metadata based on the C2PA standard. This can indicate that a file was processed by Claude and help detect whether the metadata has been tampered with.

There is an important rider though. Anthropic said a detected watermark is not conclusive proof of where content came from.

Someone may have written the original material themselves and then used Claude to proofread, translate, summarise or convert it. The resulting content can still carry a Claude mark.

Content can also change after Claude processes it. It could be edited, excerpted or combined with other material.

And the absence of a detectable mark doesn't necessarily mean AI wasn't involved.

Anthropic said a Claude-generated passage may not carry a detectable mark if it came from an older model, was heavily edited, paraphrased or translated, is too short to provide a reliable signal, or has been mixed into other writing.

For files, provenance metadata can also disappear through format conversion, re-saving or screenshots.

Anthropic is still working on the tools that will allow users and third parties to detect Claude's watermarks and provenance metadata. The company will publish more technical details on how detection works.

So this isn't quite an AI lie detector. It's more like a hidden digital trail, designed to give users and platforms another signal about whether content may have been processed by Claude.