Is AI Advertising Illegal? What You Actually Have to Disclose in 2026

Is AI Advertising Illegal? What You Actually Have to Disclose in 2026

No. But a short list of things now needs a label, and most teams cannot tell you which things.

New York started fining advertisers in June. The EU AI Act's transparency rules went binding on August 2. California's took effect the same day, aimed at somebody else entirely. Two days ago the IAB published a 42 page framework trying to hold it all together.

None of this means stop using AI. It means a few specific uses need a label and most do not. Here is which is which.

What needs a label

The IAB framework lands on seven triggers, all of them cases where somebody could reasonably be fooled.

• Images or video generated from a prompt, even after a human edits or composites them, unless the result is obviously not real.
• Photorealistic synthetic humans in a lead role.
• A synthetic voice of a dead person saying something they never said.
• A synthetic voice of a living person in a situation that never happened, as opposed to a scripted endorsement they agreed to.
• A digital twin of a dead person, in any use, even with the estate's permission.
• A digital twin of a living person in events or places that never happened.
• A chatbot inside an ad that behaves like a human.

Notice what they have in common. Every one is a person, or an image realistic enough to pass as a photograph. None of them is about efficiency, automation, or a model having touched your file.

Takeaway: No fake human and no fake photograph means you are almost certainly clear.

What does not

This list matters more, because it covers most of what you actually do with AI.

• Routine post-production. Colour correction, lighting, background cleanup, blemish removal. The test is a good one: if it would not have needed a disclosure done the traditional way, it does not need one now.
• Clearly stylized work. Illustration, obvious CGI. Nobody reads it as a photograph.
• Technical fixes. Upscaling, de-noising, format conversion.
• Copy and text. This surprises people. AI-written headlines and body copy do not trigger a consumer label.

Takeaway: The threshold is what the consumer perceives, not how much AI was in your pipeline.

The timeline

South Korea and India have added requirements too, with a grace period and draft advertising guidance respectively. If you run global campaigns, get counsel on those.

Takeaway: If you advertise in New York or the EU, the deadline already passed.

The one everyone misreads

California's AI Transparency Act applies to "covered providers," meaning generative AI systems with over a million monthly users: Adobe Firefly, DALL-E, Google's tools. It does not apply to brands.

Those providers must tuck an invisible record into everything they generate, saying it was AI-made. That happens automatically and it is already in the files you download. They must also offer a visible label, but offering it is where their duty ends. Using it is optional, and California sets no penalty on you for leaving it off.

So the hidden record is in your file whether you thought about it or not, and the part a person can actually see is entirely your decision.

Takeaway: Your vendor being compliant does not make you compliant.

The number behind the fight

An NYU Stern study found that telling people an ad was made with generative AI cut click-through by 31.5%. The IAB cites it in its own framework as the argument against blanket labeling. It is the first compliance label in advertising that arrives with a published price tag.

The more uncomfortable finding sits in the same document. 82% of advertisers think consumers feel positive about AI-generated ads. Only 45% of Gen Z and Millennial consumers do. A 37 point gap, and it has grown since 2024. The reason is in the data: cost efficiency is now the top benefit advertisers cite for AI, up from fifth place in

• The industry got excited for reasons that have nothing to do with better work, and consumers noticed.

Against that, Pew found 76% of US adults think it matters to be able to tell AI content from human content, and the IAB found clear disclosure is the third strongest driver of attention to an AI-made ad. So the label costs clicks now and buys trust later. Both are true, and no framework decides that trade for you.

Takeaway: Treat disclosure as a brand investment with a measurable short-term cost, not a compliance checkbox.

What the platforms already enforce

This moves faster than legislation. TikTok requires disclosure for anything fully AI-generated or significantly modified, and rejects or restricts ads where it finds undisclosed AI. Meta auto-labels ads made with its own tools, but third party tools do not trigger it, so declaring it yourself is the mechanism. Google stamps an invisible watermark into what its tools make, and added a "How this ad was made" panel in 2026.

Takeaway: Platforms are the real enforcement layer, and they can switch your campaign off.

The question nobody has answered

All of this rests on Content Credentials: a small invisible record tucked inside an image or video file that says how it was made. The IAB is adding two entries to that record for advertising, one saying whether a disclosure was required and one saying whether a label was actually applied. Those get wired into the plumbing that traffics ads, so platforms can check automatically instead of taking your word for it. Within a couple of years, most commercial content online will carry a hidden note about how much AI made it.

AI search engines run on exactly this kind of hidden, machine-readable signal. So does a note saying "mostly AI-made" end up shaping what gets quoted and cited in AI answers? We looked. Not one engine has published anything on it. No policy, no guidance, no statement of intent. We watched the other side of this play out when Time sold ads to AI agents and Perplexity blocked them: an engine decided what it would trust, with no published policy and no appeal.

The IAB admits the flaw in its own approach: nobody can see this record, and it is easy to remove. A screenshot wipes it. So does re-saving the file. Which is why California's January 2027 rule against stripping it out is the most consequential date on that timeline, and the one nobody is discussing.

Takeaway: The same hidden record built to label your ads is what AI engines could use to judge whether to trust you.

What this means for your brand

Your legal exposure is smaller than the noise suggests. Name one person who owns disclosure decisions, add four questions to the pre-launch checklist (was AI used, which tools, does it hit a trigger, is the label visible), and always use the same label, either the sparkle icon or plain words like "AI-generated image". That is a 60 day workflow change, not a compliance programme.

Your visibility exposure is the bigger one. The hidden record being built to label ads is the same kind of signal AI search feeds on.

If you want to know what AI engines currently say about your brand, book a meeting, or start with our research reports.

FAQ

Q: Is AI advertising illegal? A: No. Using generative AI to make ads is legal in every major market and no jurisdiction has moved to ban it. What is regulated is non-disclosure in specific cases, mainly AI-generated people and photorealistic images that could mislead a reasonable consumer. New York, the EU and several Asian markets require a label in those cases, but the underlying activity is not prohibited.

Q: Do I have to label AI-generated ad copy? A: No. The IAB framework does not treat AI-written headlines or body copy as a disclosure trigger, and the EU AI Act's text provision targets content published to inform the public on matters of public interest rather than advertising copy. Text is the biggest AI use in advertising and the least regulated.

Q: What happens if I do not disclose? A: New York sets civil penalties at $1,000 for a first violation and $5,000 for each one after. The EU AI Act carries fines up to 15 million EUR or 3% of worldwide turnover. In practice the faster consequence is platform enforcement, since TikTok rejects or restricts ads where it finds undisclosed AI content.

Q: Does California's AI Transparency Act apply to my brand? A: Not directly. It regulates "covered providers," meaning generative AI systems with over a million monthly users such as Adobe Firefly, DALL-E and Google's tools. They must add that invisible record automatically and offer a visible label. Applying that label is optional and California sets no penalty for brands that skip it.

Key takeaways

• AI advertising is legal. Undisclosed synthetic humans and photorealistic fakes are the regulated part.
• Every trigger is a fake person or a fake photograph. Copy and routine post-production are outside them.
• New York has fined advertisers directly since June 9. The EU has been binding since August
• California regulates your vendors, not you.
• Disclosure costs 31.5% of click-through now and buys trust later. The hidden record behind it may end up shaping what AI engines cite.

Sources: IAB AI Transparency and Disclosure Framework v2 (August 2026), EU AI Act Article 50, New York S.8420-A, California SB 942 as amended by AB 853, NYU Stern research on AI advertising, IAB and Sonata Insights, The AI Ad Gap Widens (January 2026), Pew Research Center.