Anthropic extends Claude text watermark to four more models, with a loophole on two

Anthropic told API customers by email that it is expanding its text watermark to four more Claude models: Claude Mythos 5, Claude Opus 4.7, Claude Opus 4.6 and Claude Sonnet 4.6. Newer models such as Claude Opus 5.5 already carry the mark, and the rollout on Amazon Bedrock may take a few extra days.

Anthropic describes the watermark as an "imperceptible pattern in word choice" that does not change meaning, quality or readability. According to the company, it adds no extra characters or tokens, does not affect pricing or request and response formats, and barely slows output speed. Anthropic also says the watermark "contains no information about the user, their organization, or their conversations with Claude."

The catch is on the two older models. On Opus 4.6 and Sonnet 4.6, Anthropic says the watermark applies only when developers leave three sampling parameters at their default values: temperature, top_p and top_k. These control how much randomness the model uses when picking words. A minor tweak to any one of them is enough to avoid the watermark. Anthropic warns in its email: "If you rely on the watermark for your own compliance, keep these settings at their defaults."

The email does not mention that caveat for Mythos 5 or Opus 4.7. The article reads this as a sign that the watermark applies to those two regardless of sampling settings, which is an inference, not a stated fact. It calls the loophole on the older models notable because Opus 4.6 remains Anthropic's best writing model, ahead of even Opus 5.5.

Anthropic points to the EU AI Act's December deadline as the reason it is watermarking older models now. Every model released after Aug. 2, 2026 shipped with the watermark on day one, while older models are getting it in stages. Because the watermark is embedded at the model level, it works on every platform that hosts these models, including Amazon Web Services, Google Cloud and Microsoft Foundry. Customers cannot turn it off. Anthropic also has a detection API in private preview for verifying the watermark.

Technically, Anthropic uses a variant of Google DeepMind's SynthID text method, which tweaks the randomness behind word selection to produce a statistically detectable pattern. Google DeepMind open-sourced SynthID for text and built it into the Gemini models. OpenAI launched its own system, textGrain, in October 2026; it activates by default only for ChatGPT and Codex users in the EU, while API customers worldwide can switch it on or off and it ships disabled.

Whether watermarking truly leaves text quality untouched is disputed. Blogger John Gruber argues that no two synonyms mean exactly the same thing, so picking words by a watermark key instead of semantic fit inevitably degrades precision. The legal trade outlet Artificial Lawyer raises a separate concern for law firms: clients who have banned AI use could discover the mark, and multiple watermarks layered inside one document could create complications.

Key facts

Anthropic is adding its text watermark to Claude Mythos 5, Opus 4.7, Opus 4.6 and Sonnet 4.6; Opus 5.5 and other newer models already have it.
On Opus 4.6 and Sonnet 4.6 the mark applies only if temperature, top_p and top_k stay at defaults, so changing any one of them avoids it.
Anthropic cites the EU AI Act's December deadline; models released after Aug. 2, 2026 shipped with the watermark from day one.
The mark sits at the model level, works on AWS, Google Cloud and Microsoft Foundry, and customers cannot turn it off.
Anthropic uses a variant of Google DeepMind's SynthID text method; OpenAI's textGrain, launched in October 2026, ships disabled for API customers.

Why it matters

Text watermarking is moving from a few vendors' experiments to a compliance feature tied to the EU AI Act's December deadline. Anthropic is retrofitting older models, not just new ones, and the gap it describes on Opus 4.6 and Sonnet 4.6 shows how easily a statistical mark can be sidestepped. The article stresses the gap because Opus 4.6 remains Anthropic's best writing model, ahead of even Opus 5.5.

Who it affects

API customers using the four models, on Anthropic's own API and on Amazon Web Services, Google Cloud and Microsoft Foundry, since the mark is built into the model and cannot be switched off. It also touches anyone who relies on the watermark for their own compliance, and professionals such as law firms whose clients restrict AI use.

How to use it

Nothing needs to be enabled; the watermark is on by the model's design. Anthropic's email advises that if you rely on the watermark for your own compliance, you should keep temperature, top_p and top_k at their defaults on Opus 4.6 and Sonnet 4.6. A detection API for verifying the mark is in private preview, and more detail is on Anthropic's content labeling information page.

How solid is it

The claims come from an email Anthropic sent to API customers, as reported by The Decoder. The claims about quality, speed and pricing are Anthropic's own and not independently tested in the article. That Mythos 5 and Opus 4.7 are free of the sampling loophole is the article's inference from the email's silence, not a stated fact. The article does not say how Opus 4.6 is judged the best writing model.

Risks and caveats

The loophole means a single parameter change can remove the mark on two models, so it is weak evidence of AI origin there. Quality is contested: John Gruber argues that choosing words by watermark key instead of semantic fit inevitably degrades precision. Artificial Lawyer warns that clients who have banned AI use could discover the mark and that multiple watermarks in one document could cause complications. The article gives no date for the email or for when the rollout finishes beyond a few extra days for Bedrock.

“If you rely on the watermark for your own compliance, keep these settings at their defaults.”

— Anthropic, in its email to API customers