D3 · CI 3003 · NoticesExplainer · 6 min

Takedowns and AI: when AI copies your work, or your work is called AI-made

Two near-identical prints side by side with a loupe, for AI-content claims
Two near-identical prints side by side with a loupe, for AI-content claims
Short answer

If someone posts AI output that copies your work, you can send a DMCA takedown to the site hosting it, as with any copy; the US Copyright Office said on May 9, 2025 that near-replica outputs likely infringe. If your own work is wrongly called AI-made, gather drafts and file history and use the platform's appeal.

AI has added two new kinds of copyright dispute. In the first, an AI system produces something that copies your work, and someone posts it. In the second, your own work is removed, rejected or flagged because someone believes a machine made it. This guide, part of our copyright notices section, covers both: when a takedown notice works for AI output, where else to report it, and how to respond when you are wrongly accused of using AI.

The pattern is not entirely new. A decade ago, the six strikes program raised similar questions about accusations based on automated detection; our archive explains why the Copyright Alert System ended in 2017 and what followed it.

01When an AI output copies your work

Most AI output does not copy any single work closely. But some does. In Part 3 of its Copyright and Artificial Intelligence report, released in pre-publication form on May 9, 2025, the US Copyright Office noted that users have shown AI systems producing near-exact replicas of film stills, recognizable characters and news text, and said: "Such outputs likely infringe the reproduction right."

What matters is the output itself. If an image or passage is substantially similar to your protected expression, it can infringe whether a person or a machine produced it. A general style, a subject or an idea is not protected, so an image "in the style of" an artist usually does not infringe on its own.

Signs an output may infringe

  • It reproduces your text word for word, or nearly so, for more than a short phrase.
  • It recreates a specific image of yours, including composition and details, not just the general look.
  • It shows your original character in a way people would recognize.

02Can you send a DMCA notice for AI output?

Yes, in the same situations as any other copy. A DMCA notice goes to the website or platform that hosts the infringing material: a social network, a marketplace, a blog host. The Copyright Office's section 512 FAQ lists what to include, and our guide to sending and answering a DMCA takedown notice walks through each step.

For AI output, a few details matter more than usual:

  • Identify your original precisely. Link to the published work, and explain which elements were copied. Reviewers may not see the similarity unless you point it out.
  • Identify the copy precisely. Give the exact URL of each post or listing.
  • Think about fair use and similarity first. A notice is a sworn statement. A loose resemblance or shared style is unlikely to support one, and false notices can create liability under section 512(f).
  • The poster may not know. People sometimes post AI images without realizing they copy someone. A takedown still applies, but a polite message may also solve it.

03Reporting to the AI tool itself

A DMCA takedown is built for material hosted online. AI output is often generated privately for one user and never posted, so there may be nothing to take down. In that case, look at the AI company's own channels:

  • Copyright or intellectual property complaint forms. Many AI companies publish a contact or form for copyright complaints, often linked from their terms of use or help center.
  • Training opt-out or removal requests. Some companies offer ways to ask that your work not be used in future training. These do not remove copies already made, and they vary widely.
  • Model or output feedback tools. Reporting a specific prompt and output that reproduces your work can help the company add guardrails.

Keep copies of what you send and any reply. If a company's model repeatedly reproduces your work, that record may matter later. Questions about whether training itself was lawful are separate from takedowns; our AI section covers training, fair use and opt-outs in detail.

04When your work is called AI-made

The reverse problem is growing: a writer's article is rejected as AI-written, an illustrator's post is removed under an AI policy, or a student's essay is flagged by a detector. Copyright rarely causes these actions directly. They usually come from a platform rule, a publisher's policy or a school's integrity code.

Why it can matter for copyright

The Copyright Office will not register material generated by AI without human authorship. Its registration guidance of March 16, 2023 requires applicants to disclose AI-generated content, and its Part 2 report of January 29, 2025 says that "prompts alone" do not make a user the author. If someone claims your work is AI-generated, they may also be suggesting that you do not hold copyright in it. Being able to show your own authorship protects both your reputation and your rights.

How to respond

  1. Ask what rule was applied and what evidence was used. A detector score alone is weak evidence.
  2. Gather your record: drafts, version history, sketches, layered source files, notes and research.
  3. Use the right process. If a DMCA notice removed your work, a counter-notice applies. If a platform removed it under its own AI policy, use that platform's appeal.
  4. Stay factual and brief, and keep copies of everything you submit.

No detector score should be the only evidence that a work is AI-made. Our guide to proving human authorship lists the kinds of evidence reviewers find most convincing.

05Keeping a record before it happens

The best time to prepare for either kind of dispute is before it starts.

  • Save drafts and use version history. Writing and design tools that keep a history make the strongest record of your process.
  • Keep dated originals. Hold on to raw photos, layered art files and early versions with their original dates.
  • Publish in a place that shows dates. A dated public post helps prove your work came first if a copy appears later.
  • Consider content credentials. Provenance labels attached to a file can record how it was made and edited. Our page on content credentials and C2PA explains what they prove and what they do not.
  • Register important works. Registration is needed before suing over a US work, and it creates a public record. Disclose any AI-generated parts.

If the message came from your internet provider rather than a platform, the steps are different; what to do about an ISP copyright notice walks through them.

Where the law is still moving

Who is responsible when an AI system reproduces a protected work, the user who prompted it or the company that built it, is not settled. The Copyright Office said in its May 2025 pre-publication report that it will address enforcement and the allocation of liability for infringing outputs in a later part of its AI report.

Sources

  1. Copyright and Artificial Intelligence, Part 3: Generative AI Training (pre-publication version)
    US Copyright Office · 9 May 2025
  2. Section 512 of Title 17: Resources on Online Service Provider Safe Harbors and Notice-and-Takedown System
    US Copyright Office · 30 October 2026
  3. Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence (88 FR 16190)
    Federal Register, US Copyright Office · 16 March 2023
  4. Copyright and Artificial Intelligence, Part 2: Copyrightability
    US Copyright Office · 29 January 2025

How we pick and check these: our editorial method.

07Questions people ask

Can I copyright my art to stop AI companies from using it?

Your art is already protected by copyright once it is made. Registration strengthens enforcement but does not by itself stop AI training. Whether training is fair use is being decided in the courts.

Can someone file a DMCA takedown against my AI-assisted work?

Only if they own a copyright your work infringes. Using AI tools is not infringement in itself. If a takedown is wrong, a counter-notice under section 512(g) is the formal response.

Is it illegal to post AI art that looks like a famous character?

It can infringe the character's copyright and sometimes a trademark, especially if you sell it. Keeping such images private is much lower risk than posting or selling them.

What should I do if a client says my work was made with AI?

Ask what the claim is based on, then share drafts, version history and source files. If your contract addressed AI use, point to its terms, and keep the conversation in writing.