Anthropic is adding an invisible mark to text from older Claude models on November 16, and customers cannot turn it off. Here is what it means for your work.
Text written with Claude now carries a hidden mark, and Anthropic is extending it to older models. Opus 4.6, Opus 4.5, Sonnet 4.5 and Haiku 4.5 get it on November 16. Three newer models, Claude Mythos 5, Opus 4.7 and Sonnet 4.6, have had it since November 3. Customers cannot switch it off.
Every Claude model released since August 2 has shipped with the mark. Now Anthropic is adding it to older ones too. Opus 4.6 was due on November 3 but was moved to November 16, with no reason given. On Amazon Bedrock, the rollout takes a few extra days.
The mark applies worldwide. It covers the Claude apps, Claude Code, Cowork and the API, which developers use to build Claude into their own tools. It also shows up in text made through AWS, Google Cloud or Microsoft Foundry.
You cannot see it. Claude slightly adjusts its word choices to create a statistical pattern that readers do not notice. The Decoder reports it is a variant of Google's SynthID. Anthropic says the mark contains no information about the user, their organization or their conversations.
The mark travels when text is copied and pasted, and it may survive some editing. Heavily rewritten, translated or very short passages may not carry a detectable mark.
Hardly anyone yet. Detection is in a private preview for regulators, law enforcement, media, fact-checkers, researchers, educational groups and companies with legal duties. Access requires an application. Anthropic also warns that a detected mark does not prove who wrote the text.
Anthropic has signed the code of practice for Article 50 of the EU AI Act, which calls for machine-readable marking of AI content. The Decoder says the company cites a December deadline. OpenAI takes a different path. Its watermark has been on by default since October, but only for ChatGPT and Codex users in the EU. API customers must turn it on themselves.
If your team uses Claude for emails or reports, assume the text carries a hidden AI mark, even if Claude only proofread it.
World models, AI that learns physics from video, are being built into home robots and driverless cars, and your own videos may help train them.
Read more: AI that learns how the physical world works is coming to robots and carsA new Alibaba-led test of real coding requests found the best AI agent fully finished just 26.4 percent of tasks, so human review still matters.
Read more: AI coding helpers fully finish only about 1 in 4 real tasksA University of Chicago law professor asked Claude to fix page numbers. It deleted his syllabus and a year of notes instead. Dropbox saved him, not the AI.
Read more: An AI assistant deleted a law professor's year of lecture notes
