Copyright, plagiarism, and attribution

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. The prompt is treated like an instruction, not authorship.

You can read the actual 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

The lawsuits making news are about how these models were *built*. Getty Images sued Stability AI, claiming millions of its photos were used for training without permission. (Some Getty watermarks even showed up in generated images.) Authors including George R.R. Martin and the *New York Times* have sued OpenAI over books and articles used as training data.

These cases are mostly about the AI companies, not you. But there is a practical takeaway: AI can sometimes spit out near-copies of real, copyrighted work, especially:

• Famous brand logos and characters (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.

Practical rules you can use tomorrow

Here is how to stay safe and honest in real work.

For text

• Treat AI as a drafter, not the author. Rewrite in your own voice. This both protects your copyright and reduces plagiarism risk.
• Run a plagiarism check on anything public-facing. Free options exist, and Google itself 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. This is the part people skip 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 / DALL·E, GPT-4o image generation): paid and free users can use output commercially under OpenAI's terms.
• Midjourney: paid subscribers get commercial use rights; free trials generally do not.
• Adobe Firefly: specifically markets itself as "commercially safe" because it was trained on Adobe Stock and licensed content. This is the safest choice for client work.

Always read the current terms before a paid project. They change.

A quick workflow for safe publishing

Here is a checklist you can copy into a doc and reuse.

Before you publish AI text:

1. Did I rewrite it substantially in my own words?

2. Did I verify every fact, stat, and quote?

3. Did I run it through a plagiarism check?

4. Does this context (school, client, publication) require disclosure?

Before you publish an AI image:

1. Does it contain any real logo, character, or recognizable person?

2. Did I name a living artist in the prompt?

3. Does my tool's license allow commercial use at my plan level?

4. Did I add my own editing or design work (which strengthens my rights)?

If you can clear all four 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 guardrail, not a lawyer. It just 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:

• Courts are slowly working through the training-data lawsuits. Some early rulings have leaned toward "training can be fair use," but the picture is far from settled.
• The Copyright Office is holding firm that pure AI output is not copyrightable, while accepting works with substantial human contribution.
• Tools are competing on provenance: showing where content came from. Adobe's "Content Credentials" attach a tamper-evident label noting AI involvement. Expect more of this.

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. Transform it meaningfully to gain rights.
• You can still be liable for the output, especially images of trademarked characters, logos, or named living artists' styles. Avoid these in published work.
• Always rewrite and fact-check AI text before publishing. This protects your copyright and prevents accidental plagiarism. Confirm every quote and citation.
• Read the tool's commercial terms. Adobe Firefly is the safest for client work; Midjourney and OpenAI grant commercial rights to paid users.
• Disclose AI use where school, journalism, or client contracts require it, and keep a simple pre-publish checklist for your team.