Claude, Copyright, Privacy, and Disclosure for Writers

Copyright still turns on human authorship

The U.S. Copyright Office has said that generative AI outputs can be protected only where a human author determines sufficient expressive elements. It also says AI assistance does not automatically bar copyright when human creativity is present.

For writers, the operational answer is recordkeeping. Keep drafts, outlines, notes, source packets, edits, and decisions that show your human contribution. The more important the work, the more important the audit trail.

Do not list Claude or Anthropic as an author. The Federal Register guidance says applicants should identify human contributions and should not name an AI technology as author or co-author simply because it was used.

Disclose AI-generated material when registration requires it

If you register a work that contains AI-generated material, the Copyright Office guidance requires disclosure and a brief explanation of the human author contribution. That does not mean every AI-assisted work is unprotectable; it means you must identify the protected human authorship accurately.

For a book, this may include human-written chapters, human selection and arrangement, human revisions, and AI-generated material excluded from the claim. For marketing work, registration may not be the main issue, but contracts and client policies can still require disclosure.

This site is not legal advice. It is a workflow guide: keep records and consult qualified counsel for publication, licensing, or registration decisions that matter commercially.

Know which Claude surface you are using

Anthropic privacy terms differ by consumer and commercial product contexts. The consumer privacy article says chats and coding sessions may be used for model improvement if the user allows it, if content is flagged for safety review, or if the user otherwise opts in. The commercial article says inputs and outputs are not used for training by default unless feedback or permission is given.

Sensitive drafts, unpublished manuscripts, client files, legal records, and regulated data deserve a policy check before upload. If you are in an organization, follow the organization plan, retention settings, and data handling rules rather than assuming a personal-account rule applies.

For personal drafting, consider whether incognito chats, privacy settings, or a commercial plan are appropriate. For organization work, ask the owner or admin what controls apply.

Memory is now part of this picture. As of September 2026 the memory used in chat is the same memory used in Claude Cowork, and Claude adds topics as you chat rather than summarizing afterward. Anthropic says memory is on by default on Free, Pro, and Max, while Team and Enterprise admins control availability and it stays off for individual users until enabled. Topics Anthropic treats as sensitive — health, race, ethnicity, religious beliefs, politics, gender identity, and similar areas — are not stored by default, and opting in applies only going forward. Everything stored is visible topic by topic in Memory settings and can be edited or deleted, which is the control to use when an unpublished project, a client name, or a source detail has been retained longer than you intended.

Use disclosure as a relationship tool

Disclosure is not one universal sentence. A publisher, school, client, newsroom, or platform may have its own rule. Some care about any AI assistance; some care only about generated text; some prohibit AI for particular submissions.

A useful disclosure is specific: AI was used for brainstorming, outline critique, copy-edit suggestions, source-table extraction, or draft generation. Specificity prevents both overclaiming and hiding.

For client work, define AI use in the contract or statement of work. For editorial submissions, read the guidelines. For public-facing content, disclose when a reasonable reader would consider AI involvement material.

• Do not fabricate testimonials, quotes, reviews, credentials, or sources.
• Do not use Claude to imitate a living writer for commercial substitution without permission.
• Do not send sensitive third-party documents into tools without authority.
• Do not let AI assistance erase the accountable human editor.

Watermarking does not make the disclosure decision for you

Since August 2026, Claude watermarks the text it generates, and Anthropic has said this is to comply with the EU AI Act after signing the EU Code of Practice on Transparency of AI-Generated Content. It is tempting to read that as the disclosure question being settled by the tooling. It is not.

Anthropic’s own framing is the clearest answer: a watermark only helps test whether Claude might have produced or processed content, and it says nothing about ownership or authorship. It carries no identifying information and cannot be traced to a person, organization, or chat. A marking obligation placed on a model provider is a different thing from an author’s duty to a publisher, client, school, contest, or reader.

So the decisions on this page stand unchanged. Copyright still turns on human authorship as the Copyright Office describes it. Registration still requires disclosing AI-generated material and identifying the human contribution. Commercial claims are still your company’s responsibility regardless of what generated the sentence. None of those obligations are satisfied, softened, or triggered by whether a watermark exists in the file.

The trap runs in both directions, which is why absence of a marking is not a defence. Text that Claude only proofread may carry no detectable watermark at all, and text output by models launched before August 2, 2026 falls under a transition period. Basing a disclosure claim on “no watermark was found” would be unsound. Base it on your records. For how the watermark works and what it can and cannot show, see the watermarking guide on this site.

• Watermarking is a provider-side marking obligation, not an author disclosure regime.
• It says nothing about ownership, authorship, or legal responsibility.
• Registration disclosure duties come from Copyright Office guidance, unchanged.
• Advertising substantiation duties come from FTC-enforced law, unchanged.
• Keep the human authorship trail as your evidence, not a detector result.