Copyright, plagiarism, and attribution
In 2022, a Colorado artist named Jason Allen won the Colorado State Fair art competition with an image he made using Midjourney. When he tried to register it with the US Copyright Office, they refused in 2023. Their reasoning: the parts a machine generated have no human author, so they cannot be copyrighted. That decision still shapes what you can and cannot do with AI output today.
This lesson gives you the practical rules for using AI text and images at work without getting sued, plagiarizing, or publishing something you have no rights to.
The big question: who owns AI output?
Two separate questions hide here. Keep them apart.
- Can you own the copyright to what the AI makes? (Usually no, or only partly.)
- Could the AI output get you in legal trouble? (Sometimes yes, even if you did nothing wrong on purpose.)
Owning AI output: the "human authorship" rule
In the US, copyright protects work created by a human. The Copyright Office has been clear: text or images generated purely from a prompt are not copyrightable. Its January 2025 report says prompts alone work like instructions, not authorship. In March 2025 a federal appeals court backed the same rule in *Thaler v. Perlmutter*, ruling that a work made entirely by an AI system cannot be registered.
You can read the guidance here: US Copyright Office report on AI and copyright.
What this means in plain terms:
- You type "a fox in a snowy forest, oil painting style" into an image tool. The raw output is not yours to copyright. A competitor could legally reuse it.
- You write a blog post entirely with ChatGPT and publish it as-is. That text has no copyright protection. Anyone can copy it.
The exception: when you add real human work
If you meaningfully edit, arrange, or build on the AI output, the parts you contributed can be protected.
Examples that usually count as human authorship:
- You generate 12 images, heavily edit one in Photoshop, combine it with your own photography, and lay it out. Your edits and arrangement are protectable.
- You use ChatGPT for a first draft, then rewrite half of it, restructure the argument, and add your own examples. Your version becomes a human work with AI assistance.
The rule of thumb: the more you transform it, the more you own it. Pure "press generate" output is fair game for anyone.
The other risk: AI output that copies someone else
Even if ownership were not an issue, AI models can reproduce material they were trained on. That creates two practical dangers.
Danger 1: The output looks too much like training data
Most lawsuits in the news are about how these models were built. Getty Images sued Stability AI over millions of photos used for training (some Getty watermarks even showed up in generated images). Authors and the *New York Times* have sued OpenAI over books and articles. In 2025 Anthropic agreed to pay $1.5 billion to settle a class action from authors whose books it had downloaded from pirate sites.
Other cases target the output itself. In June 2025 Disney and Universal sued Midjourney, showing images of Darth Vader, Shrek, and the Minions that the tool produced on request.
The practical takeaway: AI can sometimes produce near-copies of real, copyrighted work, especially:
- Famous characters and brand logos (Mickey Mouse, Pikachu, the Nike swoosh)
- Distinctive living artists' styles when you name them ("in the style of [artist]")
- Well-known photographs or song lyrics
If you publish a generated image that closely matches a real trademarked character, *you* can be the one getting the cease-and-desist letter, not the AI company.
Danger 2: Accidental plagiarism in text
Plagiarism (presenting someone else's words or ideas as your own) is a separate problem from copyright. AI text can closely paraphrase or even quote a source without telling you.
Real scenario: you ask an AI to "summarize the latest research on intermittent fasting" and it returns a paragraph lifted almost word-for-word from one blog. You publish it under your name. That is plagiarism, even though no human meant it to happen.
Who Owns AI Generated Art? The Copyright Problem Explained
Practical rules you can use tomorrow
For text
- Treat AI as a drafter, not the author. Rewrite in your own voice. This protects your copyright and reduces plagiarism risk.
- Run a plagiarism check on anything public-facing. Free options exist, and Google is a quick check: paste a suspicious sentence in quotes and search.
- Verify every fact and quote. AI invents citations. Never publish a quote you have not confirmed exists.
- Disclose AI use where it matters. Academic work, journalism, and many client contracts now require it.
For images
- Never publish AI images of trademarked characters or logos for commercial use. "A cute yellow electric mouse" is still legally risky.
- Avoid naming living artists in prompts for published work. Use descriptive terms instead: "bold flat colors, thick outlines" rather than "in the style of [artist]."
- Check your tool's license terms. People skip this part and it matters a lot (see below).
Check the tool, not just the law
Copyright law says you may not *own* the output, but the AI company's terms of service may still grant you commercial use rights. These are different things.
As of 2026:
- OpenAI (ChatGPT image generation): paid and free users can use output commercially under OpenAI's terms.
- Midjourney: paid subscribers get commercial use rights. Companies with more than $1 million in annual revenue need a Pro or Mega plan.
- Adobe Firefly: Adobe markets its own Firefly models as "commercially safe" because they were trained on Adobe Stock and licensed content, and offers IP indemnity on some enterprise plans. Watch out: the Firefly app also offers partner models from other companies, and those do not carry the same guarantee. For client work, pick an Adobe Firefly model explicitly.
Always read the current terms before a paid project. They change.
Knowledge check
1. According to the lesson, why is raw AI-generated output (created purely from a prompt) generally not protected by US copyright?
2. The lesson distinguishes two separate questions about AI output. What are they?
3. You use ChatGPT to write a first draft, then rewrite half of it, restructure the argument, and add your own examples. What does the lesson say about your resulting work?
4. Select ALL scenarios that would likely count as human authorship (and therefore be protectable) according to the lesson.
Select all the correct answers.
5. Select ALL statements that are TRUE about the practical risks of using AI output at work, based on the lesson.
Select all the correct answers.
A quick workflow for safe publishing
Here is a checklist you can copy into a doc and reuse.
Before you publish AI text:
- Did I rewrite it substantially in my own words?
- Did I verify every fact, stat, and quote?
- Did I run it through a plagiarism check?
- Does this context (school, client, publication) require disclosure?
Before you publish an AI image:
- Does it contain any real logo, character, or recognizable person?
- Did I name a living artist in the prompt?
- Does my tool's license allow commercial use at my plan level, and for the specific model I used?
- Did I add my own editing or design work (which strengthens my rights)?
If you can clear every item in each list, you are in good shape.
A small automation: flagging risky prompts
If your team generates a lot of images, you can catch the riskiest prompts automatically before anyone hits "generate." This simple Python check flags prompts that mention trademarks or named artists.
risky_terms = [
"mickey", "pikachu", "nike", "disney", "marvel",
"in the style of", "logo of", "starbucks"
]
def check_prompt(prompt):
found = [term for term in risky_terms if term in prompt.lower()]
if found:
return f"⚠️ Review needed. Risky terms: {found}"
return "✅ Looks clear, but human judgment still required."
print(check_prompt("A fox in a snowy forest, oil painting"))
print(check_prompt("Pikachu in the style of Van Gogh"))Output:
✅ Looks clear, but human judgment still required.
⚠️ Review needed. Risky terms: ['pikachu', 'in the style of']This is a guardrailguardrailRules and controls that keep an AI system inside safe, legal and on-brand boundaries, blocking outputs and actions that cross the line.View full definition →, not a lawyer. It catches the obvious mistakes before they become published mistakes.
Where the law is heading
Things are still moving in 2026. A few trends worth knowing:
- Training lawsuits are getting first answers. In 2025 two US judges (in cases against Anthropic and Meta) found that training on books could be fair use. The Anthropic judge also ruled that keeping millions of pirated books was not, which led to the $1.5 billion settlement. How the data was obtained matters.
- Outside the US, results differ. A UK court largely rejected Getty's main claims against Stability AI in November 2025, and rules in the EU and elsewhere are developing separately.
- The Copyright Office and the courts are holding firm that pure AI output is not copyrightable, while accepting works with substantial human contribution.
- Tools are competing on provenance, meaning showing where content came from. Adobe's Content Credentials attach a tamper-evident label noting AI involvement, and the EU AI Act is pushing labeling of AI-generated content.
The safe mindset: assume AI output is a raw ingredient, not a finished product you own. Your human work is what creates value and legal protection.
Key Takeaways
- Pure AI output is usually not yours to own. In the US, copyright needs a human author, and the courts confirmed this in 2025. Transform the output meaningfully to gain rights.
- You can still be liable for what you publish, especially images of trademarked characters, logos, or named living artists' styles. Disney and Universal are already suing Midjourney over exactly this.
- Always rewrite and fact-check AI text before publishing. Confirm every quote and citation.
- Read the tool's commercial terms. Adobe's own Firefly models are the safest choice for client work (partner models in the same app are not covered). Midjourney and OpenAI grant commercial rights under their terms.
- Disclose AI use where school, journalism, or client contracts require it, and keep a simple pre-publish checklist for your team.
What to do, from this lesson
These actions are compiled in the role's Playbook.
- Transform AI output meaningfully before treating it as yours
- Avoid trademarked characters, logos, and named living artists' styles in published work
- Read each tool's commercial terms before using output for clients
- Disclose AI assistance when it would change how work is trusted