Purely AI-generated content generally cannot be copyrighted in the United States or Australia, but the human parts of AI-assisted work can be. The US Copyright Office says generative AI output is protected only where a human author determined sufficient expressive elements, and that prompts alone do not give enough control; in March 2026 the Supreme Court declined to hear a challenge to the human-authorship rule. What you can own is your own expression that survives in the output, your creative selection and arrangement, and your creative modifications. The UK is the outlier: a 1988 provision still protects computer-generated works, though the government proposed removing it in March 2026.
This guide covers who owns what when a creator uses AI, and how to document it. It is general information, not legal advice. The step-by-step takedown process is in our copyright takedown evidence pack; this page explains what you can honestly claim before you file anything.
What the US Copyright Office decided
The Office's Copyright and Artificial Intelligence report was published in parts, and Part 2, released on 29 January 2025 and summarised in NewsNet issue 1060, deals with copyrightability. Its conclusions, paraphrased:
- Existing law can answer these questions; no new legislation is needed.
- Using AI to assist, rather than stand in for, human creativity does not affect protection for the result.
- Original human expression is protected even when the work also contains AI-generated material.
- Purely AI-generated material, or material where a human lacked sufficient control over the expressive elements, is not protected.
- Whether a human contribution is enough is decided case by case.
- With currently available technology, prompts alone do not provide sufficient control.
- Humans can own their own work that is perceptible in an output, plus creative selection, coordination or arrangement of outputs and creative modifications of them.
- The case for a new right protecting AI output has not been made.
The report is blunt about prompt iteration. It says repeatedly revising prompts does not change the analysis, because effort is irrelevant to copyright and each resubmission is simply “re-rolling” the dice: the output still reflects your acceptance of the system's interpretation rather than your own expression. Hours spent prompting therefore do not, on their own, earn ownership.
The courts agree on the foundation. In Thaler v. Perlmutter the appeals court in Washington, D.C. upheld the Office's refusal to register a work listing an AI system as its author, and the Supreme Court docket for case 25-449 records the petition for review as denied on 2 March 2026. A denial is not a ruling on the merits, but it leaves the human-authorship requirement in place.
What you can and cannot own in common creator workflows
The table applies the report's three routes to authorship, expressive inputs, selection and arrangement, and modification, to everyday creator work. It shows likely outcomes under US law, not guarantees; each case turns on its facts.
| Workflow | Part you are likely to own | Part you are unlikely to own |
|---|---|---|
| Typing a prompt and posting the generated image as-is | Any caption or text overlay you wrote yourself | The image, however many attempts it took |
| Feeding your own photo or drawing into an AI restyle | Your original expression that remains perceptible in the result | Elements the system added on its own |
| Cutting generated b-roll into a video you scripted and narrated | The script, your narration and the creative selection and arrangement of the edit | Each generated clip taken alone |
| Rewriting an AI first draft of a caption, script or ebook chapter | The passages you wrote or substantially rewrote | AI text you kept unchanged, beyond trivial amounts |
| Arranging AI images with your own text in a comic, lookbook or zine | The overall selection, coordination and arrangement, plus your text | The individual images |
| Hand-editing an AI image until your changes are creative in their own right | Your modifications, if they meet the originality standard | The underlying generated base |
| Posting a fully synthetic persona's daily images | Possibly a curated compilation, rarely a single post | Most single images |
The comic row reflects the report's account of an early registration decision, where the Office protected the selection and arrangement of AI images with human-written text in a comic book but not the images themselves. The persona row matters most for AI persona accounts: if most of the value sits in single generated images, much of it may sit outside copyright, which changes how you protect it.
Human-authorship evidence log
Your claim is only as good as your proof of what a human did. Keep one row per asset, filled in at the time rather than reconstructed after a dispute.
| Field | What to record |
|---|---|
| Asset ID and date | A file name or post link and the date you made it |
| Human inputs | Original photos, sketches, recordings or drafts you fed in, saved as separate files |
| Tool and settings | The AI product, version or model name, and any reference images or modes used |
| Prompts | Kept for context; the Office says they do not establish authorship by themselves |
| Raw outputs | The unedited generations you chose from, so the before state is provable |
| Human changes | What you selected, rejected, rearranged, repainted, rewrote or re-recorded, with layered or versioned project files |
| Who did it | The named person, plus a signed assignment if a freelancer or team member did the creative work |
| Final work and claim | The published file and one sentence on which parts you would claim and which you would exclude |
That last field doubles as a draft of the registration wording below and of the ownership statement in a takedown. If other people appear in or work on the content, our content rights release checklist covers the paperwork for collaborators.
US registration disclosure checklist
Australia has no copyright register; IP Australia describes protection there as free and automatic. If you do register a work with the US Copyright Office, its registration guidance for works containing AI-generated material sets a duty to disclose AI content and explain the human contribution:
- Use the Standard Application, not a simplified form.
- Name the human author or authors. Do not list an AI tool or its company as an author or co-author just because you used it.
- In the Author Created field, describe what the human contributed. For a mixed work the guidance suggests wording along the lines of selection, coordination and arrangement of the human-authored content and the AI-generated content.
- Exclude AI-generated content that is more than de minimis under Limitation of the Claim, in the Other field under Material Excluded, with a short description such as the content type followed by “generated by artificial intelligence”.
- Add context in the Note to CO field if it helps the examiner.
- If you are unsure how to fill it in, state generally that the work contains AI-generated material; the Office will contact you.
- If a pending application left out the AI material, contact the Office's Public Information Office so a note is added to the record.
- If a registration already issued without the disclosure, file a supplementary registration that describes the human contribution and disclaims the AI material.
- Do not leave it uncorrected. The guidance warns that registrants who fail to update the record risk losing the benefits of registration, that the Office may cancel a registration, and that a court can disregard one where inaccurate information was knowingly provided.
What this means when someone copies your content
A copyright takedown is a statement of ownership. YouTube's page on how to submit a copyright removal request requires you to state, under penalty of perjury, that you own or are authorised to act for the owner of an exclusive right that is allegedly infringed, and warns that misuse of the process can lead to account termination or legal consequences. Other platforms' notice forms rest on similar declarations.
- Claim the human layer precisely. For a reposted video, name the script, narration and edit you made, not the generated clips inside it.
- Expect weak footing for pure generations. If a copied image was generated wholesale, a copyright notice may not hold; impersonation, terms-of-service or privacy routes may fit better. Our impersonation report workflow covers fake accounts.
- Keep the log ready for a counter-notice. If the uploader disputes your claim, the evidence log is what shows your contribution. YouTube's dispute stages are explained in our claim versus strike guide.
- Treat likeness separately. Copyright protects works, not your face or voice. Misuse of your likeness in someone else's AI content raises different rights, and licensing your likeness deliberately is covered in our AI digital twin licence checklist.
How the UK and Australia differ
Australia. The Copyright Act 1968 does not deal with AI expressly. The Arts Law Centre of Australia's information sheet on AI and copyright explains that, as a general rule, a work is protected only if a human author contributed independent intellectual effort, so AI output without enough human input may not be protected. It suggests that editing your own image with an AI tool is likely enough, and that it remains unclear whether writing or training the algorithm, or pre-selecting and editing outputs, would be.
United Kingdom. Section 9(3) of the Copyright, Designs and Patents Act 1988 says that for a computer-generated literary, dramatic, musical or artistic work, the author is taken to be the person who made the arrangements necessary for its creation. The government's Report on Copyright and Artificial Intelligence, published on 18 March 2026, proposes removing that protection while keeping protection for AI-assisted works, and notes a contradiction in the provision that leaves its practical reach unclear. The same report points out that sound recordings, films and broadcasts are protected regardless of human creativity, so AI-generated music and video could keep some protection in that form. Until Parliament acts, the provision stays law.
If your audience and infringers are spread across countries, the place where you enforce can matter as much as where you live, which is a question for a lawyer before you rely on any of these positions.
Limitations of this guide
This summary reflects official pages and reports as they stood on 1 October 2026. The US Office has said it will supplement its registration guidance and update its Compendium, the UK proposal is not yet law, and Australian courts have not ruled on generative AI authorship, so any of this can move. The ownership table gives likely outcomes under US principles, not legal conclusions about your work, and it does not address whether a tool's training or output infringes someone else's rights, or what a platform's own terms say about AI content you upload. Platform disclosure duties are separate from copyright; for YouTube, see our YouTube AI content policy guide. Before registering, sending notices at scale or licensing a catalogue that relies on AI, get advice from an intellectual property lawyer in the relevant country.