Anthropic is putting watermarks in Claude-generated text to meet EU rules

Anthropic will watermark the text generated by its models, including Claude, to comply with the Transparency Code of the EU AI Act, the company says. The confirmation appears in an updated support page, which TechCrunch reported on.

The code, which took effect on August 2, requires AI companies to mark AI-generated or edited content in a way that other systems can identify it.

How the watermarks work

All models released after August 2 automatically get watermarking for both text and files, according to the company. For files, Anthropic is using the C2PA open standard, the same content-credentialing scheme used to prove the origin of photos and video. For text, the approach is different.

The watermark lives in the text itself. "Because the watermark is part of the text, it will travel with the text when it's copied and pasted elsewhere, and may persist through some editing," the support page reads. It is applied at the model level, so the mark is present no matter which Claude product or surface the text comes from.

The company said it will also extend watermarking to older models. It named the products where it applies: the Claude Platform API, Claude, Claude Code, Claude Cowork and Claude Tag.

One question stays open: how much editing is enough to strip the mark. The support page only says it "may persist through some editing," and TechCrunch has asked Anthropic to clarify the threshold. The report carried no answer from the company.

Part of a broader move

Anthropic is not alone in this direction. Platforms are rushing to watermark AI-generated content after user backlash and to avoid regulatory scrutiny. Last week, the AI music platform Suno said it would mark tracks created on its platform, after a run of legal challenges. Last month, the newsletter service Substack partnered with Pangram to flag AI-written content; its CEO, Chris Best, called out "Claudefishing," the term for people using AI to mass-generate content.

A longer list of companies has committed to following the EU code, including Black Forest Labs, Google, Meta, Microsoft, OpenAI and Synthesia.

What it means for you

If you consume or distribute content that may come from Claude, the mark is designed with the reader in mind: it exists so that other systems can identify the text, which is the point of the EU requirement.

Practically, two things matter if you build on the Claude API or use the products day to day. First, assume any output you ship carries the watermark. It is applied at the model level, not per product, so there is no unmarked corner of the Claude stack. Second, do not plan workflows around stripping the mark by editing, since how much editing it survives is still unconfirmed. If provenance matters in your pipeline, this is the first concrete step toward verifiable AI text output from a major lab. The claims about how well the mark holds up are the company's own for now, and the report does not mention independent testing.