# Anthropic adds its text watermark to four more Claude models

> On Opus 4.6 and Sonnet 4.6 the mark only applies at default sampling settings, The Decoder reports. Anthropic's help article lists which models carry it.

URL: https://claudegpt.si/news/anthropic-watermark-older-claude-models  
Published: 2026-10-31T22:49:02.008Z  
Author: ClaudeGPT Si Newsroom  
Section: AI News  
Topics: Anthropic, Claude, EU AI Act, AI safety  
Publisher: ClaudeGPT Si

## Key takeaways

- Anthropic is adding its text watermark to Claude Mythos 5, Opus 4.7, Opus 4.6 and Sonnet 4.6, according to The Decoder.
- On Opus 4.6 and Sonnet 4.6 the mark applies only when temperature, top_p and top_k stay at their defaults, per the same report.
- Anthropic says a detected mark means content may have been generated or processed by Claude, and a missing mark rules nothing out.

Anthropic is adding its text watermark to four more Claude models, and on two of them developers can switch it off by changing a sampling setting, [The Decoder reports](https://the-decoder.com/anthropics-watermark-rollout-leaves-a-loophole-on-two-older-claude-models/). The company's own [help article](https://support.claude.com/en/articles/16266773-how-claude-marks-ai-generated-content) confirms the EU-driven rollout and lists which models carry the mark. The Anthropic watermark is a hidden pattern in word choice that lets a detector tell that Claude handled a piece of text.

Anyone who builds on Claude and relies on the mark for compliance should read the settings caveat below.

## What is changing

According to The Decoder, which cites an email Anthropic sent to API customers, the watermark is coming to Claude Mythos 5, Opus 4.7, Opus 4.6 and Sonnet 4.6. Opus 5.5 already has it. Amazon Bedrock may take a few extra days. We could not see the email itself, so these details rest on The Decoder's account.

Anthropic's help article says it signed the EU AI Act's Article 50(2) code of practice on marking AI-generated content. Models launched in the EU on or after August 2, 2026 are marked from day one. Earlier models fall under a transition period and are being updated. The article lists Fable 5.1, Fable 5, Opus 5.5, Opus 5, Opus 4.8, Sonnet 5.5, Sonnet 5 and Haiku 5.5 as fully supported, including through cloud partners and with signed metadata on image files.

## The loophole on Opus 4.6 and Sonnet 4.6

The Decoder says that on Opus 4.6 and Sonnet 4.6 the watermark applies only when `temperature`, `top_p` and `top_k` stay at their defaults. These settings control how random the model's word choice is. Change any one and the mark is not applied. The Decoder quotes the email: "If you rely on the watermark for your own compliance, keep these settings at their defaults."

The help article does not mention these settings. The report gives no such caveat for Mythos 5 or Opus 4.7.

## What a watermark does and does not prove

Anthropic's help article is careful here. A detected mark means the content "may have been generated or processed by Claude". Claude may only have proofread or translated text that someone else wrote. A missing mark proves nothing either: short text, heavy editing, translation or an older model can all leave it out.

Detection is in private preview for eligible organizations such as regulators, researchers, media and companies with their own obligations under the Act. Anthropic says it will widen access over time.

## What to watch

The Decoder says the EU deadline for older models is in December. Anthropic says it will publish technical guidance on marking and detection as it becomes available. Developers on Opus 4.6 and Sonnet 4.6 should check whether their sampling settings change the mark before they depend on it.

## Sources

- [How Claude marks AI-generated content](https://support.claude.com/en/articles/16266773-how-claude-marks-ai-generated-content), Anthropic
- [Anthropic's watermark rollout leaves a loophole on two older Claude models](https://the-decoder.com/anthropics-watermark-rollout-leaves-a-loophole-on-two-older-claude-models/), The Decoder