Claude users will have their AI text secretly watermarked from August 2
Anthropic has started embedding invisible, machine-readable watermarks into all text generated by new Claude AI models, making the content detectable even after copying and pasting.
Anthropic has begun embedding invisible, machine-readable watermarks directly into the text generated by its new Claude AI models, a change that means any content created by the assistant will carry a hidden signal that survives copying and may persist through light editing.
The watermark is woven into the text itself at the model level, according to a support page published by Anthropic. It is imperceptible to humans, does not alter meaning or readability, and applies regardless of which Claude product generates the text—the chatbot, API, Claude Code, or deployments through cloud partners like AWS, Google Cloud, and Microsoft Foundry.
Anthropic disclosed the policy on Monday, tying it to commitments under the European Union’s AI Act Code of Practice on Transparency of AI-Generated Content. The company signed the EU code and is positioning watermarking as a global transparency measure, not one limited to European users. All Claude models launched on or after August 2, 2026 will have the watermark enabled from day one, with Anthropic working to retrofit the feature to older models as well.
Because the signal is a statistical pattern embedded during token selection—not external metadata—it travels when text is copied and pasted elsewhere. Fortune notes that heavy rewriting, paraphrasing, or running the output through another language model can degrade or remove it. Anthropic is also adding digitally signed metadata based on the C2PA open standard for non-text outputs such as images.
Critically, users cannot opt out of the watermarking. Anthropic warns that detection, while useful, is not foolproof: a detected mark only signals that content was processed by Claude—it does not prove Claude authored the text, since users routinely employ the tool for proofreading, summarising, or translation of human-written material. The company plans to publish detection tools but cautions they will not be fully conclusive under heavy editing or file-format changes.
The global rollout automatically includes African startups, developers, and enterprises that build on Claude, whether directly via API or through hyperscalers. For Nigerian media, fintechs, and regulators watching the EU’s lead on AI governance, the system offers a concrete example of technical enforcement of transparency rules.
