Anthropic is rolling out its text watermarking more broadly, meaning text will carry an invisible trail to show it was written with Claude, Andrew Griffin reported on Mon, 2 November 2026. In August, Anthropic announced it would integrate a text watermark into its models, an invisible pattern inside the words that can be spotted by an automated system, launched in response to EU regulation requiring such technology across generative AI tools, but initially only in Anthropic's newest models. Now the company said it will be added to older systems such as Mythos 5, Opus 4.7, Opus 4.6, and Sonnet 4.6. There is no way to opt out and it will cover all users globally, despite being a response to European regulation; the markings are entirely visible and there is no way for normal users to check whether their writing would be flagged. Only eligible organisations as defined under EU law – such as regulators, law enforcement, journalists and researchers – can actually detect the watermark, and it is also available to companies that have to verify watermarking under the EU's AI Act. The August announcement brought criticism from users who said it was unfair, including that proofreading or fact-checking text with Claude could trigger detection and that bad actors could re-phrase work with another AI system to hide the watermark. Anthropic accepted some weaknesses, including that shorter text might not include enough of the pattern to be spotted, but pushed back against criticism that the