A new layer of traceability will follow Claude’s words across platforms, even after copying and some editing, changing how AI-written content is identified.
Anthropic will introduce watermarks for texts generated by its artificial intelligence models, including Claude. The company is implementing labeling in line with the European Code of Practice on transparency under the AI Act, which came into force on August 2.
European requirements stipulate that AI developers must label generated or edited content in a way that allows other systems to automatically determine its origin. Anthropic confirmed the launch of this feature in an updated support section.
Which Anthropic texts will be watermarked
The labeling technology will be automatically added to all Anthropic models released after August 2. It will apply to both text content and files. For files, the company will use the open C2PA standard.
Anthropic plans to expand watermark support to models released earlier as well.
Because the watermark is part of the text, it will move with the text when it is copied and pasted elsewhere, and may remain after some editing. Labeling will be applied at the model level, meaning it will be present regardless of which Claude product or interface the text comes from.
– on Anthropic’s support page
The company did not specify how much editing would be required for the watermark to disappear. Labeling will apply to materials created through the Claude platform API, as well as in Claude, Claude Code, Claude Cowork, and Claude Tag services.
AI Act requirements for AI-generated content
Technology companies are increasingly introducing tools to identify materials created by artificial intelligence. Such solutions are influenced by regulatory requirements as well as user demand for transparency regarding the origin of texts, files, and other materials.
Black Forest Labs, Google, Meta, Microsoft, OpenAI, and Synthesia have also joined the European Code of Practice on transparency. The introduction of watermarks is intended to simplify the automatic detection of AI-generated content and help distinguish it from materials created by humans.