# Anthropic sets Nov. 16 watermark rollout for four Claude models

> The watermark is scheduled for four Claude models, but API users can bypass it by changing sampling settings. Amazon Bedrock may get the update later.

Oossa · 2026-11-04 · https://oossa.com/en/anthropic-sets-nov-16-watermark-rollout-for-four-claude-models

Event date: 2026-11-04

Anthropic has told API customers it plans to add text watermarking to Claude Opus 4.6, Opus 4.5, Sonnet 4.5 and Haiku 4.5 starting November 16, 2026. The Decoder reports that Opus 4.6 was meant to receive the watermark on November 3, but Anthropic confirmed the delay without giving a reason. The update may reach Amazon Bedrock a few days after the main rollout.

The watermark is not a visible label. Anthropic describes it as an “imperceptible pattern in word choice” that can be detected statistically. The company says it does not add characters or tokens, change meaning or readability, affect pricing, or include information about users or their conversations.

## When will the watermark apply?

For all four models in this rollout, the watermark applies only when three sampling settings—temperature, top_p and top_k—are left at their defaults. These settings affect how much variation a model uses when choosing words. Anthropic says changing any one of them is enough to avoid the watermark, so the mark is not guaranteed if developers adjust those settings.

That is a practical caveat for teams that want the mark to help with compliance: Anthropic advises keeping the settings at their defaults if they rely on it. The company says customers cannot switch the watermark off directly, but the sampling restriction means API users can prevent it from applying on these models. Anthropic also has a watermark-detection API in private preview.

## Why is Anthropic adding it to older models?

Anthropic points to the EU AI Act’s December deadline. It says models released after August 2, 2026, shipped with watermarking, while older models are being updated in stages. Because the mark is built into the model, Anthropic says it works on hosting services including Amazon Web Services, Google Cloud and Microsoft Foundry.

This rollout follows an earlier planned wave covering Claude Mythos 5, Opus 4.7, Opus 4.6 and Sonnet 4.6. The Decoder says the latest customer email does not mention Mythos 5, Opus 4.7 or Sonnet 4.6, whose November 3 rollout had been planned; whether those models received the watermark as scheduled is not confirmed in the update.

## The facts

- Anthropic scheduled watermarking for Claude Opus 4.6, Opus 4.5, Sonnet 4.5 and Haiku 4.5 to begin November 16, 2026.
- The watermark applies to these four models only when temperature, top_p and top_k are all set to their defaults.
- Anthropic says Amazon Bedrock may receive the rollout a few extra days later.
- Anthropic says models released after August 2, 2026, shipped with watermarking.

## Why it matters

For developers using these Claude models, changing a sampling setting can mean their output no longer carries Anthropic’s watermark, which matters if they planned to rely on it for compliance or content checks. Anthropic has a detection API in private preview, but the update does not say when it will be broadly available.

## Sources & references

1. [Anthropic's text watermark is easy to bypass on several older Claude models](https://the-decoder.com/anthropics-text-watermark-is-easy-to-bypass-on-several-older-claude-models/) – The Decoder, 2026-11-04

Last updated: 2026-11-04