Who Owns Your AI Ads? Copyright, Liability and the EU AI Act in 2026

On 26 September 2026 a creator on X posted a 45-second cartoon called Smooth Operator: a man, an umbrella, a rainy city. The creator says openly that it was made with AI. By the creator's own count it passed 10 million views about 32 hours later. Three days after it went up, another artist redrew the character by hand and posted it with four words: "this is mine now".

Half joke, half protest, and a fair question for any brand running AI-made ads. If a character, a voice or a whole ad came from prompts alone, who owns it? Often, nobody. This guide covers that gap, who pays when an AI ad goes wrong, and the EU labelling rules that have applied since 2 August 2026. It is the written companion to the 12-minute film above, with every source linked, for brands spending $10K a month or more on ads.

Jump to the part of the film you need:

• 1:31 Human authorship: no author, no owner
• 3:03 "This is mine now": the viral AI cartoon
• 4:40 Who pays when an AI ad goes wrong
• 5:41 Different rules in every country
• 8:21 A compliance-first strategy

Prompts alone don't make you the owner

In January 2025 the U.S. Copyright Office published Part 2 of its report on copyright and AI, on whether AI outputs can be protected at all. Its conclusion is the rule every AI ad now lives under:

The outputs of generative AI can be protected by copyright only where a human author has determined sufficient expressive elements. This can include situations where a human-authored work is perceptible in an AI output, or a human makes creative arrangements or modifications of the output, but not the mere provision of prompts.

The courts agree. In Thaler v. Perlmutter, a work made by an AI system with no human author was refused registration, and on 2 March 2026 the U.S. Supreme Court declined to hear the appeal (docket No. 25-449). The human-authorship rule stands.

For an advertiser, that has three practical effects:

• A fully generated visual, voice or video may have no copyright owner. If a competitor reuses it, there may be nothing to enforce.
• A tool's terms cannot fix that. Many tools say the output is yours. They can only hand over rights that exist, and where no human shaped the work there may be no copyright to hand over.
• The human parts are protectable. The Copyright Office also confirms that using AI to assist, or including AI material inside a larger human-made work, does not bar protection. A script your team wrote, a character finished by hand, your own footage, and the way a human selects, arranges and edits the pieces can all carry protection.

Which is the twist in the cartoon story. The redraw is human work, so it can be protected. The original, if it came from prompts alone, cannot. The person who copied may hold more than the person who created. Now picture that character as your mascot, or your best-performing ad.

When an AI ad goes wrong, the bill comes to you

Ownership is half the problem. The other half is liability: if an AI ad borrows someone's likeness, a protected design or a competitor's work, who answers for it? Read the platform's terms. Under Meta's Commercial Terms, if anyone brings a claim against Meta related to your content or your use of its products, "you agree to indemnify and hold us harmless": you cover the damages, losses and expenses, including reasonable legal fees. In plain terms, a claim about your ad is yours to pay, including the platform's lawyers. Check the equivalent clause on every platform you buy media from.

In the US there is also a rule aimed squarely at AI-made social proof. The FTC's Consumer Reviews and Testimonials Rule, in force since 21 October 2024, bans reviews and testimonials that misrepresent that they come from someone who does not exist, "such as AI-generated fake reviews", or from someone with no real experience of the product. Courts can impose civil penalties for knowing violations. The FTC's own Q&A adds that the rule has no blanket ban on AI avatars in marketing. The line is deception: an AI presenter explaining your product is allowed; an AI "customer" describing results they never had is not.

The EU rules that apply now

Article 50 of the EU AI Act has applied since 2 August 2026. It splits the work between the companies that build AI tools (providers) and the businesses that use them (deployers). The European Commission's FAQ on Article 50 is explicit that an advertising company using an AI system is a deployer, even when freelancers or contractors operate the tool on its behalf.

For advertisers, the duty that matters most is labelling deepfakes. Under Article 3(60) a deepfake is AI-generated or manipulated image, audio or video that resembles existing persons, objects, places, entities or events and would falsely appear to be authentic or truthful. The Commission's guidance reads that as three tests at once: close resemblance, a subject that exists or plausibly could, and a real chance the audience takes it as authentic. When all three are met:

• disclose it clearly, at the latest on first exposure, with a visible or audible label people notice without special tools;
• do not rely on the tool's hidden machine-readable mark, which the Commission says is not enough on its own;
• for evidently creative, satirical or fictional work, a lighter disclosure that does not spoil the work is enough.

AI-generated text published to inform the public on matters of public interest must also be labelled, unless a person with editorial responsibility reviewed it. Most product copy is not that kind of text; a health or finance explainer may be.

Two details catch non-EU brands out. The Act covers deployers outside the EU where the output is used in the EU (Article 2(1)(c)), so a brand based in the US or the UK that runs ads to EU audiences is in scope. And the Digital Omnibus on AI (Regulation (EU) 2026/1744, in force since 27 July 2026) postponed the high-risk rules to December 2027 but left Article 50 on its date. It only gave tool providers whose systems were already on the market until 2 December 2026 to add machine-readable marks. Your labelling duty did not move. Content made before 2 August 2026 does not need labelling retroactively, though the Commission encourages it.

The fines are set in Article 99 of the AI Act. Breaching the transparency rules can cost up to €15 million or 3% of worldwide annual turnover, whichever is higher; for SMEs and start-ups the cap is whichever is lower. The headline figure, €35 million or 7%, is reserved for banned practices such as manipulative AI that distorts people's decisions and causes significant harm.

A compliance-first workflow for AI ads, by budget

None of this means you stop using AI. It means five habits, applied the same way every time:

1. Assume you disclose. Label anything a viewer could take for a real person, place or event, and every AI presenter.
2. Use tools whose terms you have read. Prefer tools that allow commercial use and say what their models were trained on.
3. Keep a human in the loop who changes the work. A human-written script and human edits are what create protectable authorship, and they catch the errors a model will not.
4. Get watermark-ready. Keep the provider's marks and Content Credentials intact through your edit and export.
5. Document everything. Tool and version, prompts, who edited what, and the label decision for each market.

How much process that needs depends on your spend:

• Under $10K a month. One checklist per ad: tool, label, consent for any real face or voice. Keep the source files.
• $10K to $75K a month. A named person signs off each AI ad before launch. Keep a log per creative and give EU-bound campaigns the stricter label by default.
• $75K a month and up, or agencies. Write the policy down. Get creators and freelancers to assign rights in their human contributions by contract, review each platform's terms every quarter, and send any ad using a real person's likeness or voice to a lawyer first.

What to skip: five mistakes that cost the most

1. Reading "you own the output" as copyright

A tool's terms decide what the tool claims, not what the law protects. If the ad matters, make sure a human shaped it.

2. AI "customers" in testimonials

An avatar presenting your product is fine. An avatar presented as a buyer describing results is the fake testimonial the FTC rule bans.

3. Cloning a real face or voice without written consent

Many US states protect a person's name, image and voice, and in the EU a realistic likeness is a deepfake that must be labelled. Get consent in writing, and label it.

4. Treating the hidden watermark as your label

The Commission says deployers cannot rely on the provider's machine-readable mark to meet their disclosure duty. The label has to be something a person sees or hears.

5. Waiting for a takedown to start your records

The time to log tools, prompts and edits is when the ad is made. After a complaint, the brand with records answers in a day; the brand without them has to pull the campaign while it reconstructs what happened.

Want to know where your current ads stand? Run the free AI & Compliance Risk Check. It reads your site and your answers against fixed rules and sends you a red, amber and green checklist with the fix for each item, so you can start with the riskiest ad first. If you would rather walk through it, book a free Fix-It Call.

This guide is general information, not legal advice. For a decision about a specific ad, speak to a lawyer in each market where it runs.

Sources

• U.S. Copyright Office, NewsNet 1060: Part 2 of the AI Report, 29 January 2025.
• Supreme Court of the United States, Thaler v. Perlmutter, No. 25-449: petition denied 2 March 2026.
• Meta, Commercial Terms, indemnification clause.
• Federal Trade Commission, final rule banning fake reviews and testimonials (August 2024) and questions and answers.
• European Commission, Transparency obligations under Article 50 of the AI Act, updated 24 July 2026.
• Regulation (EU) 2024/1689 (the AI Act), Articles 2, 3(60), 50, 99 and 113, and Regulation (EU) 2026/1744 (Digital Omnibus on AI).
• The Smooth Operator posts on X: the original, the creator's 10 million count, the creator on AI art and the redraw.