POPA ReconAnthropic plans to insert an invisible watermark into all text generated by its new Claude model in the future.
Anthropic plans to insert an invisible watermark into all text generated by its new Claude model in the future. This watermark is not metadata but included in the text itself, so it follows along even if you copy and paste it elsewhere and may remain even after some editing. Applicable to models released after August 2, 2026, measures in accordance with EU AI Act-related regulations signed by Anthropic We are currently working on a way to apply the watermark to the Claude model, and the scope of application is all Claude-available regions worldwide. https://support.claude.com/en/articles/…
The excerpt is quoted from a public Telegram channel, which owns its text.