Policy & safety · 1 November 2026
Anthropic expands watermark to four Claude models, leaves bypass on Opus 4.6 and Sonnet 4.6
Anthropic will watermark Claude Mythos 5, Claude Opus 4.7, Claude Opus 4.6 and Claude Sonnet 4.6 to meet the EU AI Act’s December deadline. On the older Opus 4.6 and Sonnet 4.6, developers can bypass the mark by tweaking a single sampling parameter. The watermark is embedded at the model level, works on all cloud platforms and cannot be turned off; newer models such as Opus 5.5 already carry it.
Why it matters
Claude Opus 4.6 and Sonnet 4.6 can bypass the watermark via a sampling tweak; plan feature messaging around immutable watermark expectations.
For product managers
For your product
What does this mean for your product?
Try it on your product More from this issueAll AI news from 1 Nov →
Get this every morning, written for your product.
Deltaday reads the whole AI news cycle and tells you what changed, whether it's real yet, and what to do about it.