Crypto
Home›Crypto›Market Structure›Anthropic embeds invisible watermarks in newer Claude…
Anthropic embeds invisible watermarks in newer Claude AI outputs
The watermarking started with Claude models launched in the EU on or after August 2, 2026 and is designed to persist when text is copied, pasted, and possibly edited.
Anthropic has begun embedding an imperceptible, machine readable watermark into every piece of text generated by its newest Claude AI models, with the change rolling out first for models launched in the EU on or after August 2, 2026, and then expanding worldwide.
According to Decrypt, the watermark is applied at the model level and is intended to reach all Claude surfaces, including the chatbot, the API, Claude Code, and cloud partners such as AWS, Google Cloud, and Microsoft Foundry. Anthropic says users will not see the mark, and it should not change the meaning, quality, or readability of responses.
Decrypt reports that Anthropic also uses signed metadata under the C2PA open standard, effectively adding a second layer that records who produced a file and whether it was altered. Anthropic has not publicly detailed how the watermark is created, and it has not yet released detection documentation or the exact technique.
Open source projects aimed at removing the watermark appeared within days, and Decrypt notes researchers are already exploring approaches to break the watermarking scheme. Some experts suspect the method relies on a statistical signature that subtly shifts word choices, though that remains a hypothesis until Anthropic publishes more information.