Two questions arrive with anything AI makes for you: was the model allowed to learn from other people’s work, and is the new thing yours to own? The law is still arguing about both. The most useful skill is telling a ruling from a rule, sorting every claim into one of three buckets so a single court’s decision never gets mistaken for settled law.
If you remember one thing: Copyright is a bargain, not a wall, a limited deal meant to keep both people and machines making good things. The big training question is honestly unsettled as of July 2026, and no ruling yet has answered it for good. What a prompt hands back is probably no one’s; what you shape with your own hands can be yours.
| Bucket | What it means | How to state it |
|---|
| Decided ruling, with limits | A real decision, but from a single trial court on its own facts | Report it as a fact of record, name the limits (trial-court level, fact-specific, appealable or already on appeal); never downgrade it to merely “contested” |
| Settlement, no merits inference | A deal to end a case | Report the deal as a fact; a settlement is not a court finding, and settling is not an admission of anything |
| Genuinely open question | Still being fought over, or never litigated | Say so plainly, with an “as of July 2026” hedge; never upgrade it to “decided” in either direction |
| Factor | What it asks |
|---|
| Purpose of the use | Does it add something genuinely new (transformative, weighs toward fair use) or just repackage the original? |
| Nature of the work | Plain facts get thinner protection; creative work gets more |
| Amount used | Taking a little weighs toward fair use; taking the whole thing weighs against it |
| Effect on the market | Often the decider: does the new use compete with the original and eat into its income? |
Whether training AI on copyrighted work is fair use is a genuinely open question as of July 2026. Courts have ruled in narrow, fact-specific, conflicting cases, and no higher court has settled it. Do not let anyone tell you it has been answered, in either direction, neither that training is fair use nor that it was ruled infringement.
| Situation | Can it be copyrighted? |
|---|
| Purely AI-made, no human author | No. Settled at the top for this clear situation only: human authorship is required |
| The bare output of a single prompt | Probably not yours; typing prompts alone does not steer the expression closely enough |
| Your creative selection, arrangement, and editing of AI output | Yes, that human part can be protected, even when the raw AI-made part cannot |
| Your own writing that shows up in the result | Yes, that human-authored expression can be protected |
The more of you in the final piece, the more there is to protect. This is the Copyright Office’s registration practice, its own practice today, not a final word from the courts.
| Allowed | Barred |
|---|
| Imitating a general style (a song in the style of an artist you admire) | Copying a specific protected work |
| A work merely in a similar style is not a derivative work | A derivative work, built directly on a specific protected work, needs the owner’s permission |
A separate law, the right of publicity, protects a person’s name, voice, and likeness, so a voice clone could run into that even when it does not break copyright.
Every entry is re-verified live before publish and on each freshness sweep, because this ground moves fast.
| Case | Bucket | Status and limit |
|---|
| Thomson Reuters v. Ross | Decided, with limits | Delaware court, February 2025, a startup’s copying not fair use; the tool was not a generative AI, the decision turned on its own facts, and it has since been argued on appeal with a decision still to come |
| Bartz v. Anthropic (ruling) | Decided, with limits | California court, June 2025, training on lawfully-bought books was fair use and highly transformative, while keeping pirated copies was infringement; the training holding bound only the named plaintiffs |
| Bartz v. Anthropic (settlement) | Settlement, no inference | Anthropic agreed to pay at least 1.5 billion dollars and destroy its pirated copies; covers past conduct only; went before a court for review in 2026; a settlement is not an admission |
| Kadrey v. Meta | Decided, with limits | California court, June 2025, ruled for the AI company on training, but the judge stressed the plaintiffs argued poorly and said in most cases such training is likely infringing |
| Thaler v. Perlmutter | Decided, with limits | Appeals court affirmed in 2025 that an author must be human; the Supreme Court declined the case in 2026; limited to a fully-autonomous work with no human involvement |
| New York Times v. OpenAI and Microsoft | Genuinely open | Still in discovery, much allowed forward in 2025, no ruling on the core question, no trial date, no settlement as of July 2026 |
| Allen v. Perlmutter | Genuinely open | The AI-assisted line: an artist’s refused registration for a prize-winning AI-assisted image, still being fought in court, no final ruling as of July 2026 |
| Andersen v. Stability AI and Getty v. Stability AI (US) | Genuinely open | Visual artists and an image library, active pretrial in California, no merits ruling as of July 2026 |
| Getty v. Stability AI (UK) | Non-US, non-binding | A separate case under a different country’s law reached a decision in late 2025 that mostly went against the image library; it does not bind United States courts and is itself being appealed |
| Pitfall | Correction |
|---|
| Treating one court’s decision as the law of the land | A single trial court, ruling one case on its own facts with an appeal pending, is a data point, not a nationwide verdict |
| Reading a settlement as a confession | A company can settle to make a risk go away and admit nothing; settling is not an admission |
| Assuming the prompt makes it yours | Typing a request is not authoring a work; what you can own is the creative human part you added, not the bare output |
| Mistaking style for a copy | Sounding like an artist is generally allowed; reproducing their specific work is not |
| Line | Meaning |
|---|
| Copyright is a bargain, not a wall | A limited, temporary deal meant to encourage more creation, not a permanent fence |
| Tell a ruling from a rule | One court deciding one case on its own facts is not settled law |
| The training question is honestly open | No ruling yet has answered it for good, in either direction, as of July 2026 |
| Know your split | The bare output of a prompt is probably no one’s; the creative work you add can be yours |
| Style is not a copy | Imitating a general style is generally allowed; reproducing a specific work is not |
