Claude adds hidden watermark to 4 more models

Anthropic will add a hidden watermark to text from 4 more of its Claude AI models, The Decoder reported on the 31st of October 2026. The models are Claude Mythos 5, Opus 4.7, Opus 4.6 and Sonnet 4.6.

Anthropic told customers in an email, The Decoder said. Amazon's Bedrock cloud may get it a few days later.

What the watermark is

The watermark is a hidden pattern in the words Claude picks. Readers cannot see it, and Anthropic says it does not change meaning or quality.

The mark holds no information about the user, Anthropic said, according to The Decoder. It only signals that Claude may have produced the text.

Why older models get it now

Anthropic signed a European Union code on labelling AI content under the EU AI Act, its help page says. The Decoder said a December deadline in the law is the reason.

Who can check for it

For now, only some groups can test text for the mark, Anthropic says. They include regulators, police, media, fact checkers and researchers.

A gap on 2 models

On Opus 4.6 and Sonnet 4.6, the mark works only at default settings, The Decoder reported. A small change to the settings that control word randomness switches it off.

Anthropic's email told customers who rely on the mark to keep those settings at their defaults, The Decoder said.

Below we break down the model list

Anthropic's help page lists 16 Claude models in its watermark table. The 4 new models are one quarter of that list, by our arithmetic.

What is confirmed and what is not

The help page confirms the watermark plan and the model list. The start date for the 4 models did not appear on the pages we opened.

Key facts

Anthropic will watermark text from 4 more Claude models, The Decoder said.

They are Claude Mythos 5, Opus 4.7, Opus 4.6 and Sonnet 4.6.

The mark cannot be seen and holds no user data, Anthropic said.

On Opus 4.6 and Sonnet 4.6 it works only at default settings.