Who Owns What AI Writes? Copyright Basics for Office Workers
Date Published
You are allowed to use what AI writes for you, and you almost certainly do not own it. Those are two different things, and most office workers collapse them into one. ChatGPT, Claude and Gemini all hand you the right to use their output commercially. United States copyright law, confirmed by a federal appeals court in 2025 and left standing by the Supreme Court in 2026, says a work with no human author gets no copyright at all. The practical consequence is blunt: a brochure, product description or marketing image that AI produced on its own can be copied by a competitor, and you have no claim. What you can protect is the part a person actually made. This article is general information, not legal advice; the rules differ by country and the edges are still being argued in court.
What did the courts actually decide?
That an AI system cannot be an author, and the question is now settled in the United States for the foreseeable future. In Thaler v. Perlmutter, an inventor tried to register an image his AI system had generated with no human involvement, naming the machine as the author. The DC Circuit affirmed the refusal on 18 March 2025, reading the Copyright Act's references to an author's lifespan, heirs and signature as things only a human can have. The Supreme Court declined to hear the appeal on 2 March 2026, which leaves the appeals court ruling as the law. The court was careful about one thing: it ruled only on a work where the machine was the sole author. Work made by a person using AI as a tool was explicitly left open, and that is where the useful part of this story lives.
So what do the AI companies give me when I use their tools?
Permission, and whatever ownership exists, which may be none. Read the wording in OpenAI's terms of use slowly: the company assigns to you "all our right, title, and interest, if any, in and to Output." The two words that matter are "if any." OpenAI is not promising that a copyright exists; it is promising that if one does, it is yours rather than theirs. The same terms also warn that output may not be unique and that other users may receive similar results. Anthropic and Google take the same general position for their consumer and business products. So the tool gives you a licence to use and a disclaimer about ownership in the same breath, and the disclaimer is the honest part.
What does this mean for work I produce with AI?
Anything AI made on its own is free for anyone to reuse, including your competitors. If you prompt an image generator for a hero picture and drop it on your website unchanged, a rival can lift that exact image and you have no infringement claim, because there is nothing to infringe. The same applies to a product description you copied straight out of a chat window and to a slide deck the assistant wrote from a one line brief. This is not theoretical; the Copyright Office has already refused or limited registrations for AI images on exactly this basis. For internal documents nobody wants to copy, it does not matter. For anything customer facing that carries your brand, it matters a great deal.
How much human work makes it mine again?
Enough that a person controlled the expression, not just the idea. The US Copyright Office's January 2025 report on copyrightability draws the line in a way an office worker can act on. Prompts alone are not enough, no matter how long or detailed, because a prompt expresses an idea while the machine decides how that idea is expressed. What does count: substantially editing or rewriting the output, selecting and arranging generated pieces into an original whole, feeding in your own creative work that stays visible in the result, or using AI as one tool inside a process you direct. The protection covers the human contribution, not the raw machine output underneath it, and the Office decides case by case.
What should I do differently at work?
Treat AI output as raw material and keep evidence that you shaped it. Rewrite anything customer facing in your own words rather than lightly polishing it; a heavy edit is protectable, a spell check is not. Keep the drafts, because a version history showing the machine's first pass and your changes is the record you would need if ownership were ever questioned. For images, prefer AI as a starting point that a designer then composes, crops, combines or paints over, and never let a logo be pure AI output. Logos are protected mainly by trademark, which cares about whether customers recognise the mark, not who drew it, so register the mark properly and keep the design work human. If you use AI images for anything visible, the workflow in creating images with AI for work already assumes a human finishing step for exactly this reason.
Is there a separate risk that AI output copies someone else?
Yes, and it is a different problem from ownership. Ownership asks whether you can stop others copying you. Infringement asks whether your output copies someone else, because models are trained on existing work and can reproduce it closely. Well known characters, distinctive art styles tied to a living artist, song lyrics and long passages of text are where this bites. Keep AI output away from anything that recognisably echoes a specific creator, and if a piece is going on a billboard rather than an intranet, have someone check it. Our guide on reviewing contracts with AI covers the same habit of verifying before you rely.
Does this change outside the United States?
Yes, so check your own jurisdiction before assuming. The United Kingdom has long had a provision giving computer generated works a shorter term of protection, with authorship attributed to the person who made the arrangements for the work to be created, though the government has been consulting on whether to keep it. The European Union follows the same human authorship principle as the United States in substance, and the EU AI Act adds transparency duties around AI generated content rather than granting ownership. Multinational teams should assume the strictest reading and write a line about this into their team AI usage policy, so nobody has to guess.
The tool lets you use what it makes. Only your own work lets you own it.
Pick the one customer facing piece your team relies on most that AI drafted. Rewrite it properly, keep both versions, and you have turned something anyone could copy into something that is yours.
Frequently asked questions
Can I use ChatGPT output commercially?
Yes. OpenAI, Anthropic and Google all permit commercial use of output under their terms. Permission to use is separate from owning a copyright, which under US law requires a human author.
Can I copyright something AI wrote for me?
Not the raw output. You can register the parts a human contributed: substantial edits, creative selection and arrangement, or your own work incorporated into the result. Prompting alone does not count.
What happened in Thaler v. Perlmutter?
A US appeals court ruled in March 2025 that an AI system cannot be an author under the Copyright Act, and the Supreme Court declined to review that ruling in March 2026. The case only covered works with no human author.
Can a competitor copy my AI generated marketing image?
If the image was pure AI output with no meaningful human authorship, there is no copyright to enforce, so in practice yes. Human editing, composition or combination can create a protectable version.
Is an AI generated logo safe to use?
Use is allowed, but keep the final design human made and rely on trademark registration, which protects a mark based on recognition rather than who drew it. This is general information, not legal advice.
Sources
End the stock photo era: describe images like a design brief, keep one reusable style recipe so everything matches, and inspect hands and text before shipping.
Research shows disclosing AI use can cost you trust, and hiding it costs more. A practical spectrum for when to mention it, when not to, and how to say it well.