As of 2026-11-12, Anthropic's updated Usage Policy is in force, and its disclosure duties are specific rather than general: the policy text contains no rule that every Claude-written text must carry a label. It does bar submitting AI-assisted work without proper permission or attribution, requires consumer chatbots to say they are AI, and requires people to be told when AI produced high-risk advice or a decision about them.
What changed and when
Anthropic announced the 2026 Usage Policy update on 2026-10-08 and set 2026-11-12 as the effective date; the policy page is dated "Nov 12, 2026". MIXED and PCWorld both reported the same date on 2026-10-09.
The announcement lists these changes:
- One section on deceptive activity, political or commercial, including fake accounts and fabricated news sites.
- An elections section narrowed to deceiving voters and disrupting elections, with the blanket ban on personalised vote targeting removed.
- Clearer rules on weapons, and a rewritten surveillance and law-enforcement section.
- Continued human oversight and notice to affected people for high-risk uses.
- A new ban on sustained and needless abusive behaviour toward Claude itself.
The announcement does not describe a new labelling rule for AI-written text, and it does not mention academic use.
What the policy says about disclosure and attribution
| Topic | Where in the policy | Wording |
|---|---|---|
| Attribution | Fraudulent, Abusive, or Predatory Practices | Using Anthropic's products to "Plagiarize or submit AI-assisted work without proper permission or attribution" is listed as prohibited |
| Consumer chatbots | Additional Use Case Guidelines | Must "disclose to users that they are interacting with AI rather than a human"; the disclosure need not name Anthropic, Claude or the model |
| High-risk advice | High-risk Use Case Requirements | A person who receives advice or is the subject of a decision "must be clearly told that AI was used to produce it" |
| Concealed sponsorship | Deceptive Campaigns or Artificial Activity | Prohibits misrepresenting or concealing the sponsorship of content meant to influence public opinion, officials or decision-makers |
The table covers what Anthropic's page states. It is not a summary of the whole policy.
What it means for teachers, editors and developers
Teachers. The attribution line is a term of Anthropic's own policy for people who use its products. It does not define misconduct at your school. Your course rules and your institution's process decide that, and Turnitin's own statements on what it covers are on the Turnitin and Claude page.
Editors. Nothing in the policy requires a byline note on Claude-assisted copy. The lines it draws are concealed sponsorship, impersonation and fake outlets. Whether a publication labels AI use is a house-policy decision.
Developers. If you ship a consumer chatbot on Claude, tell users it is AI; you need not say it is Claude. Anthropic's support article adds that anyone deploying Claude in their own product should independently assess what Article 50 of the EU AI Act requires of them.
Does the policy change how Claude text is detected?
No. Anthropic's watermark and its detection API are a separate matter: the watermark is in private preview for eligible organisations, and individuals cannot check text themselves. The watermark status page has the dated details. Style-based detection does not depend on any of it.
What the policy does is make undisclosed use a possible issue where permission or attribution is expected. That raises the stakes of a check, and a check gives evidence, not a verdict. A detector flag is a reason to ask the author a question.
Check a text
Paste at least 100 words into the Claude AI detector. It shows whether the text reads as AI-written, highlights the passages that carry the signal, and lists writing-style similarity to Claude and other model families. For fully AI-written text from the Anthropic family, Veread, Fileverity's text model, detects 98.4%, and it wrongly flags 0.5% of human-written texts in seven languages; the facts page has the details.
The result cannot tell you who wrote a text, whether permission was given, or whether the author should have disclosed their use. Light edits and short texts weaken any detector. For the phrase-level tendencies of Claude's writing, see how to tell if Claude wrote it.
Sources, all read on 2026-11-12: Anthropic, 2026 Usage Policy update; Anthropic Usage Policy; Claude Help Center, How Claude marks AI-generated content; MIXED, 2026-10-09; PCWorld, 2026-10-09.