AI and copyright
This is general information, not legal advice — but there's one specific myth worth clearing up before anything else, because it causes real problems.
The myth that trips up the most people
Paying for a Midjourney or ChatGPT subscription feels like it should mean you own what it produces. It doesn't, at least not automatically. A paid subscription gives a commercial license — permission to use the output — which is a different thing entirely from copyright ownership, which is about who legally authored the work. Confusing the two is the single most common, and most consequential, mistake in this whole area.
Who actually owns AI-generated content
Under current US law, content that's purely AI-generated — typed in a prompt, took the first output with no meaningful changes — cannot be copyrighted by anyone. The U.S. Copyright Office has consistently required human authorship, and that standard was reinforced when the Supreme Court declined, on March 2, 2026, to hear a challenge over an AI-generated artwork submitted for registration with no human author listed. The lower court's reasoning stands: a work with no human author has no one who can hold its copyright. It isn't that the AI owns it — nobody does.
What actually can be protected
The moment a person substantially edits, arranges, curates, or builds meaningfully on an AI output, the human-authored parts of that result can qualify for copyright — just not the purely AI-generated elements underneath them. A photograph with a small AI-extended background stays protected as a whole; an image that's almost entirely AI-generated with minor human tweaks is a much harder case, and exactly where that line sits is still being actively worked out by courts.
Can AI companies use copyrighted material to train models?
This is the other half of the copyright picture, and it's genuinely unresolved. Multiple major lawsuits — including publishers and artists against several large AI companies — are actively working through whether training an AI model on copyrighted material counts as fair use or as infringement. No appellate court had definitively settled this question as of early 2026. Anyone relying on a firm answer here is relying on a moving target.
What responsibility falls on the person using AI
The risks worth actually tracking
Unregistrable work
- Purely AI-generated content that can't be protected at all
Voided registration
- Failing to disclose AI use, even unintentionally
Infringement exposure
- Output that echoes existing copyrighted work too closely
How creators can actually protect AI-assisted work
Document the human decisions along the way — prompts, edits, drafts, what got changed and why. Make deliberate, substantive creative choices rather than accepting a first output wholesale. When registering, disclose AI involvement specifically and describe which elements were human-authored. The more visible the human creative contribution, the stronger the claim to protection over that specific contribution.
Where this is still moving
The human-authorship requirement for registration is now fairly settled after the 2026 Supreme Court decision. What "significant enough" human contribution actually means in edge cases is still being worked out case by case, and the training-data fair use question remains genuinely open, with major litigation ongoing. Treat this area as evolving, not fixed — what's accurate today may shift as more of these cases get decided.
The short version
A subscription buys permission to use an output, not authorship over it — and purely AI-generated content currently has no copyright owner at all under US law. Protection follows genuine human creative contribution: real edits, real arrangement, real judgment, documented and disclosed honestly. None of this is legal advice for a specific situation — for anything with real commercial or legal stakes, that's a conversation for an actual intellectual property attorney, not a general guide.