Skip to content
Catalyst Note CN markCNCatalystNote

Noting the changes that set others in motion.

Model updates

Anthropic adds text watermark to four Claude models ahead of EU deadline

Anthropic is embedding detectable patterns in four Claude models, though two older versions allow API users to bypass the mark by adjusting settings.

Anthropic adds text watermark to four Claude models ahead of EU deadline
Depth
1 outlet · newest source 376h ago

Key points

  • Anthropic is watermarking Claude Mythos 5, Opus 4.7, Opus 4.6, and Sonnet 4.6.
  • Users can disable the watermark on Opus 4.6 and Sonnet 4.6 by changing sampling parameters.
  • The rollout aims to meet the EU AI Act’s December compliance deadline.

Anthropic announced it is expanding its text watermark to four additional Claude models, including two that currently allow users to bypass the detection mechanism. The move addresses regulatory requirements while introducing a technical inconsistency across the model portfolio.

The claim

According to The Decoder, Anthropic is applying the watermark to Claude Mythos 5, Claude Opus 4.7, Claude Opus 4.6, and Claude Sonnet 4.6. Newer models like Claude Opus 5.5 already carry the mark. The rollout on Amazon Bedrock may take a few extra days. The company describes the watermark as an imperceptible pattern in word choice that does not change meaning, quality, or readability. It adds no extra characters or tokens, does not affect pricing or request formats, and barely slows output speed. The watermark contains no information about the user, their organization, or their conversations.

Evidence: reportedAnthropic's watermark rollout leaves a loophole on two older Claude models

aifleet-decoder

Evidence: reportedAnthropic's watermark rollout leaves a loophole on two older Claude models

aifleet-decoder

The response

Anthropic applies the watermark only at default settings on Claude Opus 4.6 and Claude Sonnet 4.6. Users can dodge the mark by tweaking any one of three sampling parameters: temperature, top_p, and top_k. These parameters control randomness in word selection. Anthropic warns that if users rely on the watermark for compliance, they must keep these settings at defaults. The email does not mention this caveat for Claude Mythos 5 or Claude Opus 4.7, suggesting the watermark applies regardless of sampling settings on those models.

Evidence: reportedAnthropic's watermark rollout leaves a loophole on two older Claude models

aifleet-decoder

Evidence: reportedAnthropic's watermark rollout leaves a loophole on two older Claude models

aifleet-decoder

Why this matters now

The Decoder notes that Opus 4.6 remains Anthropic’s best writing model, ahead of Opus 5.5. This makes the bypass on that specific version notable for enterprise users. Anthropic points to the EU AI Act’s December deadline as the reason for watermarking older models now. Models released after Aug. 2, 2026, shipped with the watermark on day one, while older models are getting it in stages. The watermark is embedded at the model level, working on every hosting platform, including AWS, Google Cloud, and Microsoft Foundry. Customers cannot turn it off.

Evidence: reportedAnthropic's watermark rollout leaves a loophole on two older Claude models

aifleet-decoder

Evidence: reportedAnthropic's watermark rollout leaves a loophole on two older Claude models

aifleet-decoder

What buyers should weigh

Buyers must assess whether the bypass on Opus 4.6 and Sonnet 4.6 undermines their compliance strategy. If an organization relies on the watermark to demonstrate adherence to the EU AI Act, using non-default sampling parameters on these two models negates that evidence. The cost implication is neutral, as Anthropic states the watermark adds no tokens or pricing changes. Vendor choice may shift toward Mythos 5 or Opus 4.7 for users who want guaranteed watermarking regardless of parameter tweaks. Market position is affected as Anthropic retrofits older models to meet regulatory deadlines, ensuring parity across its portfolio.

Evidence: reportedAnthropic's watermark rollout leaves a loophole on two older Claude models

aifleet-decoder

Evidence: reportedAnthropic's watermark rollout leaves a loophole on two older Claude models

aifleet-decoder

Why it matters

  • Enterprises using Opus 4.6 for high-quality writing face a compliance gap if they adjust sampling parameters for creative control.
  • The inability to disable the watermark across all platforms ensures consistent regulatory signaling for buyers using multi-cloud deployments.
  • The distinction between models that allow bypass and those that do not creates a fragmented compliance landscape for the same vendor.
  • The upcoming EU AI Act deadline forces immediate action on legacy models, accelerating the adoption of content labeling standards.

Topics: Anthropic, Claude, EU AI Act, watermarking, compliance

Updated

Ask the article

Local keyword search of this article; no generated answers.

Did this article give you the evidence you needed?

No vote saved on this device.
Your choice stays on this device. No public tally.

Questions readers ask

Can I turn off the watermark on Claude Opus 4.6?

Yes, but only by changing the default sampling parameters like temperature, top_p, or top_k, which may affect output randomness.

Does the watermark cost extra?

No, Anthropic says the watermark adds no extra tokens, does not affect pricing, and barely slows output speed.

Which models are getting the watermark now?

Claude Mythos 5, Claude Opus 4.7, Claude Opus 4.6, and Claude Sonnet 4.6 are receiving the watermark update.

Connected stories

This storyAnthropic Max Plan Now Includes $100 Monthly API Credits For Claude Platform BuildersAnthropic Max Plan Now Includes Anthropic signs first AP news license to power live election results in ClaudeAnthropic signs first AP news liOpus 5.5 safety rules halt benign RE tasks and game recreations per user reportsOpus 5.5 safety rules halt benigOpen-weight AI models match closed systems in simulated network intrusionsOpen-weight AI models match clos

Reading receipt

  • Anthropic is watermarking Claude Mythos 5, Opus 4.7, Opus 4.6, and Sonnet 4.6.
  • Users can disable the watermark on Opus 4.6 and Sonnet 4.6 by changing sampling parameters.
  • The rollout aims to meet the EU AI Act’s December compliance deadline.
Time spent: 0:00 · Depth: Full

Sources

  1. The Decoder

Related stories

Background reading