Who owns humanized AI text? Copyright, vendor terms and disclosure

Only the human parts. The Copyright Office's Part 2 report on copyrightability, released 29 January 2025, "concludes that the outputs of generative AI can be protected by copyright only where a human author has determined sufficient expressive elements", which "can include situations where a human-authored work is perceptible in an AI output, or a human makes creative arrangements or modifications of the output, but not the mere provision of prompts" (US Copyright Office, Copyright Office Releases Part 2 of Artificial Intelligence Report, read 9 November 2026). The same release states that "the use of AI to assist in the process of creation or the inclusion of AI-generated material in a larger human-generated work does not bar copyrightability".

Apply that to the two ways people use a humanizer. A writer who drafts a text and runs it through a humanizer still has a human-authored work perceptible in the output, and the human contribution is what the Office says remains protectable. A writer who prompts a chatbot, pastes the result into a humanizer and submits what comes back has supplied two prompts and no expression; on the Office's reasoning, neither machine step produced protectable authorship, and the reworded output is not more the writer's for having been reworded.

The Office's registration guidance of 16 March 2023 asks applicants to disclose AI-generated material in a work and to describe the human contribution (US Copyright Office, Copyright and Artificial Intelligence, read 9 November 2026). A registration that omits the machine steps is the applicant's risk, not the vendor's.

The two sets of terms read for this guide say they do not, and one says plainly that the output may not be anyone's. WriteHuman's Terms of Service, last updated 13 October 2026, state: "to the extent we have any rights in them, we assign to you our rights in the Outputs we deliver to you", permit use "for any lawful purpose, including commercial purposes", and add that "other users may receive Outputs that are the same as or similar to yours, and those Outputs are not yours" (WriteHuman, Terms of Service, read 9 November 2026). The same section states: "We make no promise that any Output is original, accurate, free of errors, protectable by copyright, or free from third-party rights."

StealthGPT's Terms of Use, last updated 20 October 2026, state that the company does "not assert any ownership over your Contributions" and that users "retain full ownership" of them, and separately forbid using any text "generated, rewritten, humanized, returned, or otherwise processed by the Services" to train, benchmark or build a detector for AI models, read 9 November 2026. Ownership of the output sits with the user as far as the vendor is concerned; what the user may do with it is restricted in one direction.

An assignment of "our rights, to the extent we have any" is the vendor saying it cannot give what it does not hold. Read beside the Copyright Office's position, the clause means the vendor steps aside and the question of whether anything protectable exists is left where it was. The other ten vendors in the launch set have terms pages that were not read for this guide; each tool's review links its terms, and the privacy guide covers what the same pages say about storing and training on your text.

Ownership is a national copyright question in each member state and is outside this page. Disclosure is not. Article 50 of the EU AI Act requires deployers of an AI system that generates or manipulates text "which is published with the purpose of informing the public on matters of public interest" to disclose that the text has been artificially generated or manipulated, with an exception where the content "has undergone a process of human review or editorial control" and a person holds editorial responsibility (Regulation (EU) 2024/1689, Article 50(4), EUR-Lex, read 9 November 2026). Paragraph 2 of the same article requires providers of systems that generate text to mark outputs as artificially generated in a machine-readable form, with an exception for systems that "do not substantially alter the input data".

A humanizer is a system that manipulates text, and its purpose is to alter the input. The paragraph 2 exception for systems that do not substantially alter input data is written for spell-checkers, not for a tool whose selling point is how much it changes. Whether a particular humanizer vendor marks its output under Article 50(2) is a question for the vendor's compliance statements, and none of the pages checked on 8 or 9 November 2026 mentions the article.

Three things, in order. First, settle whether the draft was yours, because everything above turns on that: a human draft reworded by a machine keeps a human author; a machine draft reworded by a machine has none. Second, read the vendor's terms for the output clause and for any restriction, as WriteHuman and StealthGPT each carry one, and keep a dated copy. Third, disclose where a rule requires it: Article 50 for public-interest text in the EU, the Copyright Office's guidance for a US registration, and a school's or publisher's own policy everywhere else, which the cheating guide covers for students.

No detector score should be the only evidence in a misconduct case, and no vendor's assignment clause is evidence that a text is yours to register. This site's own terms state how its test data may be cited and what a score does and does not mean. This page is not legal advice; the primary sources are linked above, and the rest of this series is in the guides hub.