Anthropic watermarks Claude AI content

Anthropic has begun embedding invisible, machine-readable watermarks into text generated by its Claude models, and adding digitally signed provenance metadata to supported files, as part of its compliance with the European Union's AI Act. The watermarking applies to all Claude models launched on or after August 2, 2026, and is being rolled out globally, not just in the EU. The text watermark is woven into the statistical pattern of word choices, making it invisible to humans but detectable by tools Anthropic plans to publish. It can survive copying and light editing, but may be removed by heavy rewriting or translation. For files, Anthropic uses the C2PA standard to attach signed metadata to formats like PNG, JPG, and SVG. The feature covers the Claude API, Claude web and mobile apps, Claude Code, Claude Cowork, and Claude Tag, as well as deployments on Amazon — Amazon Web Services, Alphabet Inc. — Google Cloud Platform, and Microsoft. Anthropic signed the EU AI Act's Article 50(2) Code of Practice on Transparency of AI-Generated Content. The company cautions that the presence or absence of a watermark is not definitive proof of AI authorship, as Claude may be used for proofreading or summarizing human-written text, and older models or heavy editing can remove marks. Anthropic is not the first to implement such watermarking; Alphabet Inc.'s SynthID-Text is a more mature approach, and Alphabet Inc. has partnered with Apple, ElevenLabs, Kakao, Nvidia, and OpenAI for interoperable watermarking. Anthropic's move puts pressure on other labs that signed the same EU code but have not yet shipped working features.