RIP Claude
This title could be clearer and more informative.Try out Clickbait Shieldfor free (5 uses left this month).
Anthropic announced that Claude models born on or after August 2, 2026 will watermark generated text and images, with older models retrofitted by December 2, 2026. There is no opt-out at any plan tier, watermarking is applied server-side worldwide, and no public detector exists—only Anthropic can check. Actions like synthesis, proofreading, translation, summarization, and critique all trigger watermarking, which the author, a longtime Claude Max subscriber who uses Claude for editing and research on books, calls writer-hostile. The author is building safeguards against accidentally injecting watermarked text into their writing, researching alternatives to Claude for writing tasks, and considering canceling their Claude Max subscription, while stating they have no issue with watermarking applied to Claude Code's generated code.
Questions this post answers
Will Claude watermark text it generates for tasks like proofreading or summarization?
Yes, Anthropic's watermarking applies to text produced through synthesis, proofreading, translation, summarization, and critique, not just full generation. Models born on or after August 2, 2026 will have watermarking built in, and older models are being retrofitted by December 2, 2026, with no per-model disclosure of when the switch happens. Writers weighing AI tools for editing work can track changes like this on daily.dev before they get caught off guard.
Can I opt out of Claude's AI text watermarking?
No, there is no opt-out available at any subscription or API tier, including enterprise plans. The watermarking is applied server-side, below the model level, worldwide, and no public detector currently exists to check for it—only Anthropic can verify whether text is watermarked. Developers deciding whether to keep relying on Claude for writing tasks can follow policy shifts like this on daily.dev.
Does Claude Code's generated code get watermarked the same way as Claude's writing output?
Yes, Claude Code's output is subject to the same watermarking policy as any other Claude-generated content, since the marking is applied server-side across all Claude usage. One user noted this doesn't bother them for code since they don't claim authorship of AI-generated code, unlike prose they write themselves. Developers comparing AI coding assistants and their content policies can keep up with changes like this on daily.dev.