Nov 2, 2026
Four more Claude models get an invisible text watermark
Anthropic is adding its invisible text watermark to Claude Mythos 5, Opus 4.7, Opus 4.6 and Sonnet 4.6 from Nov 3. No user ID, can't be turned off, detection in private preview. Reason: EU AI Act Article 50.
Sir Toastimir
Sources · Claude Help Center · The Decoder
Anthropic told API customers that four more models get Claude's invisible text watermark starting Nov 3: Claude Mythos 5, Opus 4.7, Opus 4.6 and Sonnet 4.6. Amazon Bedrock follows a few days later. Models launched on or after Aug 2, 2026 have carried it from day one.
What it is (and isn't)
- An imperceptible pattern in word choice. Anthropic says it doesn't change meaning or quality.
- Anthropic says it contains no information about the user, their organization or their conversations.
- It's applied at the model level, on every surface (Claude apps, API, Claude Code, AWS, Google Cloud, Microsoft Foundry). Customers can't turn it off.
Who can detect it
Detection is a private preview for eligible groups: regulators, law enforcement, media, fact-checkers, independent researchers, educational organizations, EU civil society groups, and enterprises obligated to verify it under the law.
The loophole
Per Anthropic's email (reported by The Decoder), on Opus 4.6 and Sonnet 4.6 the watermark only applies at default sampling settings (temperature, top_p, top_k). Heavily edited, translated or very short text may also not be detected.
Why
Anthropic signed the EU AI Act Article 50 Code of Practice on transparency of AI-generated content.
Should AI text always be detectable?
Links
Get your daily slice of AI
One short email with the best new AI tools, robots and AI & tech stock news. Free. Unsubscribe anytime.
We only use your email to send the newsletter.