Anthropic will begin embedding invisible watermarks into text generated by its Claude large language model and chatbot in a move designed to satisfy new European Union rules on AI transparency.
“Claude models launched on or after August 2, 2026 support marking at launch. We’re also working to add marking support to Claude models released before that date,” the San Francisco based AI giant said in a statement.
The new marking policy includes all of the services provided by Claude, including its consumer app, the developer-facing Claude Platform (API), Claude Code, Claude Cowork and Claude Tag, as well as versions of Claude accessed through AWS, Google Cloud and Microsoft Foundry.
The watermark will either be weaved in as “an imperceptible watermark directly into the text itself” or as metadata, so that when “Claude generates a supported file type, such as a .svg, .png, or .jpg, it will attach signed provenance metadata… [that] signals that a file was processed by Claude and lets you detect whether the file has been tampered with.”
Anthropic said the watermark would apply to every region where Claude is offered, not just the EU.
The marking will not be limited to Europe. Anthropic said the watermark would apply everywhere Claude is offered worldwide, another example of how the EU’s standard of compliance has ripple effects on the global standards being set for AI.
How the watermark works
Anthropic uses two separate techniques. The first includes inserting an imperceptible pattern directly into generated text, invisible to readers but detectable by machines and able to travel with the text even after it has been copied and pasted elsewhere.
According to the company, it doesn’t change the meaning, quality or readability of Claude’s response.
The second technique applies to files. It functions more like a digital paper trail, similar to EXIF data on a photograph, and is designed to be checkable rather than concealed.
Seeing it in practice would likely require a C2PA-aware tool or Anthropic’s planned detection system, rather than a standard “properties” menu, and it would not necessarily identify Claude by name so much as flag that the file passed through an AI system.
The company has also been explicit that the metadata can be lost.
Anthropic said a file’s marking can be stripped through format conversion, re-saving, screenshots, or other similar processes, meaning a document or image that once carried a Claude mark may no longer show any trace of it after being edited or converted.
Why now
The timing traces directly to the EU’s AI Act.
The act requires that AI systems creating synthetic content, such as deepfakes, mark their outputs as artificially generated under Article 50.
Regulators expect providers to rely on a multi-layered marking strategy, combining digitally signed metadata, imperceptible watermarking and, in some cases, fingerprinting or logging as a fallback, since the Code of Practice makes clear that no single marking technique is sufficient on its own.
The transparency obligations under Article 50 took effect on 2 August 2026. Non-compliance can trigger fines of up to €15 million or 3% of total global annual turnover, whichever is higher.
Can people still cover their tracks?
Anthropic has been explicit that a detected watermark is a signal, not proof.
A mark showing up in a piece of text confirms only that it may have passed through Claude at some point, not the full history of how that text came to be.
People often use the assistant to proofread, translate or summarise material that originated elsewhere, so a watermark can appear on text Claude did not originally write.
The reverse is also true: heavy editing, translation, very short passages, or use of an older model without marking support can all mean genuine Claude output goes undetected.
Anthropic is not alone in this. Google DeepMind has already developed SynthID for marking AI-generated text, images and audio, while OpenAI has discussed watermarking approaches but has moved more slowly to deploy them broadly.