Anthropic Will Add Hidden Watermarks to Text and Files From Its AI, Claude

Anthropic will start adding hidden watermarks to text and files created by its AI models, including Claude, to follow new European Union rules. The company confirmed the change on August 11, 2026, through an updated support page (TechCrunch).

The EU's AI Act Transparency Code took effect on August 2, 2026. It requires AI companies to label content created or edited by AI in a way that computers can detect, even though people cannot see it. All Anthropic models released after that date will include the watermarking automatically, covering both text output and generated files. Models launched before August 2 are still being updated to add the feature (India Today). Anthropic has said it will bring watermarking to its older models as well.

For files, Anthropic is using the C2PA open standard, a set of rules backed by a group of media and technology companies. You can think of C2PA as a digital label attached to a file that records where it came from and whether AI helped create it. The text watermark works differently: it is built into the AI model itself, so it appears no matter which Claude product produces the text. According to Anthropic's support documentation, the watermark stays with the text when it is copied and pasted somewhere else and may even survive some editing.

The watermarking will apply across all of Anthropic's products, including the Claude platform API, Claude, Claude Code, Claude Cowork, and Claude Tag.

Anthropic is not the only company falling in line with the EU's transparency rules. Black Forest Labs, Google, Meta, Microsoft, OpenAI, and Synthesia have all committed to following the EU's code of practice on transparency of AI-generated content (European Commission).

For developers who use Claude's API to build their own products, there is an important practical detail. The watermark will be present in the output by default. That means if a developer takes text from Claude's API and reshapes it for their own app, the hidden label could still be there. Developers should not assume that reformatting or re-transmitting model output will remove the watermark.

There is some history worth noting here. In July 2023, OpenAI, Google, and other companies pledged at a White House event to develop watermarking systems for AI-generated content across text, images, audio, and video (Reuters). That was a voluntary promise made in a very different regulatory environment, and it produced little in the way of lasting, shared watermarking tools. The EU AI Act changes the picture by turning what was voluntary into a legal requirement. Whether the specific watermarking methods Anthropic is using will hold up when someone actively tries to remove or fool them is still an open question.

Anthropic's Transparency Hub, updated on July 23, 2026, lists technical watermarking of AI-generated content among the company's voluntary commitments related to public awareness of AI capabilities and limitations (Anthropic). The watermarking announcement comes during a busy period for Anthropic. The company announced Claude Sonnet 5 on June 30, 2026, and Claude Opus 5 on July 24, 2026, both focused on improvements in coding, agentic workflows, and professional work at scale (Anthropic News). On July 1, 2026, Anthropic announced the global return of Fable 5 alongside a proposed industry-wide framework for scoring jailbreak severity, developed with Amazon, Microsoft, Google, and other Glasswing partners. On July 9, 2026, the company launched a public initiative asking for difficult questions about AI and committing to transparent responses.

What this creates, in the most practical sense, is a hidden label that travels with AI-generated content rather than depending on the company that made the AI to disclose it separately. For the broader ecosystem, it raises a testable question: can a watermark added during generation survive the messy reality of how text actually moves through systems? The answer will determine whether this approach becomes a lasting standard or just a box checked for compliance.