Anthropic will start applying its text watermark to Claude Mythos 5, Claude Opus 4.7, Claude Opus 4.6 and Claude Sonnet 4.6 on 3 November 2026, according to a customer email shared on X on 1 November. The email says rollout on Amazon Bedrock “may take a few additional days”.
By our count of Anthropic’s Help Center table, updated this week, 7 of the 16 Claude models it lists carry the text watermark only on Anthropic’s own surfaces, with no mark listed for AWS, Google Cloud or Microsoft Foundry and no C2PA file credentials. The other 9 carry all three.
Business Pill · The Model and the Harness
A short explainer on the difference between an AI model and the product built around it. It teaches the general idea only and says nothing about the watermark in this piece.
The key insight: As we read Anthropic’s Help Center table, the watermark is moving down the model list in stages: by our count, 9 of 16 listed Claude models carry all three marks, and 7 carry the text watermark only on Anthropic’s own surfaces.
The email, headed “Text watermarking is coming to more Claude models”, went to organisations that use one or more of the four models on the Claude Platform. It says: “On November 3, 2026, Anthropic will begin applying its text watermark to outputs from these four models”.
It adds that “This is the same watermark our newer models, including Claude Opus 5.5, already carry.” Anthropic’s release notes for 1 September say of Claude Fable 5.1 and Claude Mythos 5.1 that “Marking requires no changes to your requests or response handling.”
Which marks each Claude model carries, per Anthropic’s Help Center table (updated this week): 9 of 16 listed models carry all three; 7 carry the text watermark only on Anthropic’s own surfaces. Counts are ours. Bold names are the four models in Anthropic’s customer email, watermarked from 3 November 2026.
How the Watermark Works
Anthropic explained the method in an August post. It says “Nothing is added to the text and there are no hidden characters”, and that the watermark “carries no identifying information and can’t be traced to a specific person, organization, or chat”.
The pattern sits in word choice. Where two words would be equally good, the choice is settled by a key instead of an arbitrary random number, so text can later be checked against that key. Anthropic says the technique comes from Google DeepMind’s SynthID-Text paper.
On removal, the post says “Light editing probably won’t remove the watermark completely; a complete rewrite where every word is replaced will.” It also says “A watermark can only determine that Claude was likely involved with the content at some point.”
Who Can Check It
Detection is not open to everyone. Anthropic says its detection API is “in private preview” for “eligible organizations as required under EU law (such as regulators, law enforcement, media, fact-checkers, independent researchers, educational organizations, and EU civil society groups)”, and for enterprises that must verify watermarking for their own compliance.
Why Now: Article 50
Anthropic says it is watermarking “to comply with the EU AI Act”, and that it signed the EU Code of Practice on Transparency of AI-Generated Content in July 2026 along with “around 190 total signatories”. Article 50(2) of the AI Act requires that outputs are “marked in a machine-readable format and detectable as artificially generated or manipulated”.
Anthropic also says it applies the watermark globally because “we don’t yet have a durable way to scope it by region”. For models launched before 2 August 2026, Anthropic says, “The EU law includes a transition period”, and it is “working to add watermarking for those models as well”.
The Business Lever: Provenance Becomes a Platform Feature
As we read the customer email, the four models it names are added on the Claude Platform first, on 3 November, while rollout on Amazon Bedrock may take a few additional days.
As we read Anthropic’s August post, the watermark sits in word choice rather than in added characters, so it travels with copied text and, Anthropic says, may persist through some editing.
As we read the access rules, the ability to check for the watermark is narrower than the watermark itself: detection is in private preview for regulators, media, researchers and similar groups, and for enterprises obligated under the Act.
Anthropic, How Claude’s text watermarking works
“Watermarking carries no identifying information and canโt be traced to a specific person, organization, or chat”
Three Implications
FOR COMPANIES WRITING WITH CLAUDE Anthropic says the mark carries no identifying information about the user, their organisation or their chats.
FOR BEDROCK CUSTOMERS The customer email says rollout on Amazon Bedrock may take a few additional days after 3 November.
FOR REGULATORS AND NEWSROOMS Anthropic says detection is in private preview for eligible organisations under EU law, including media and fact-checkers.
The framework says “The governance layer sits at the intersection of capability, regulation, national security, access control, and geopolitical alignment.”
As we read it, a watermark that only approved organisations can check is one example of regulation and access control meeting in the same product decision. The framework predates the watermark rollout and does not discuss it.
We have seen the customer email only as a screenshot posted on X; Anthropic has not published it. The Help Center table does not give dates per model, and it lists no cloud-partner mark for the four models, while the email says rollout on Amazon Bedrock may take a few additional days. The table also lists Opus 4.5, Sonnet 4.5 and Haiku 4.5 with the same marks; the email does not mention them.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growthstory. This week it started being priced as a credit story.
Anthropic will apply its text watermark to four older Claude models from 3 November 2026, its customer email says. By our count of its Help Center table, 7 of 16 listed Claude models carry the watermark only on Anthropic’s own surfaces, and detection is in private preview for regulators, media, researchers, similar groups and obligated enterprises.
A note on sourcing. The 3 November date and the four model names come from an Anthropic customer email that we saw as a screenshot posted on X; Anthropic has not published it. How the watermark works, who can detect it and why Anthropic is applying it come from Anthropic’s August post and its Help Center article; the model counts are our own count of the Help Center table. The Article 50(2) quote is from the text of the EU AI Act.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
