Anthropic is watermarking Claude's text output, starting in the EU
New Claude models launched in the EU from August 2, 2026 embed an invisible, detectable pattern in generated text, Anthropic's answer to the EU AI Act's AI-content-labeling requirement.
The EU AI Act's transparency obligations for general-purpose models with systemic risk started being enforced on August 2, 2026. The European Commission can request information, evaluate a model directly, order mitigations, restrict availability, or fine a company up to 3% of its worldwide annual turnover. One obligation is marking AI-generated content as such. For new Claude models shipped in the EU from that date, Anthropic's answer is a watermark built into the text.
How it works
Anthropic is using SynthID Text, a method Google DeepMind published in 2024. It adds no visible markup. It steers Claude's low-stakes word choices (picking "overcast" over "grey," for instance) into a pattern a reader can't see but anyone holding the matching key can recover. A detection API is coming. Anthropic states the limits: light editing survives, but "a complete rewrite where every word is replaced" removes it, at which point it's fair to ask whether the output is still meaningfully AI-generated. Code is barely affected, since functional correctness leaves little room for the arbitrary word choice the watermark depends on. If it shows up anywhere in code, it's in a comment.
What governance teams should take from it
This is a compliance answer to a legal requirement. It is not a misuse-detection product: enough rewriting strips it, and Anthropic says so. Whether tracing text back to a model matters for your organization depends on what you use generated text for. Decide that explicitly before treating watermarking as either solved or irrelevant.