When You Must Disclose AI Content: The 2026 Law and Rules

You must disclose AI-generated content in three main situations: when EU law requires it from 2 August 2026 under the AI Act, when undisclosed AI content would mislead a consumer (a live FTC and UK advertising-law risk now), and when a platform's own rules demand a label. Miss any of the three and you're exposed to regulators, civil claims, or losing reach and monetisation. The rules don't ban synthetic content. They ban hiding it. This is a guide to who has to disclose what, and when.

The trigger is realism. When AI output was obviously artificial, like a cartoon, nobody needed a label. Now a generated video, voice, or testimonial can pass for real, so the law has caught up. Here's the part most teams get wrong: there's no single global rule. There are overlapping regimes, and one campaign can sit under all of them at once.

When does the EU AI Act require you to disclose AI content?

Article 50 of the EU AI Act is the big one, and it bites from 2 August 2026. It splits the duty between two roles.

If you provide the AI system (the model maker), you have to mark synthetic audio, image, video, or text so it's "detectable as artificially generated or manipulated" in a "machine-readable format." Think embedded metadata or a watermark, not just a visible caption. The Act asks for solutions that are "effective, interoperable, robust and reliable as far as this is technically feasible." (EU AI Act, Article 50)

If you deploy the system (the brand using it), the duties land on you in two cases:

• Deepfakes. If you publish AI content that resembles real people, places, or events, you must disclose that it's been artificially generated or manipulated. Artistic, satirical, or fictional work gets a lighter touch: you flag that synthetic content exists, in a way that doesn't ruin the piece.
• AI text on public-interest matters. If you publish AI-generated text to inform the public on matters of public interest, you disclose it, unless a human reviewed it and a person or body holds editorial responsibility.

Timing matters. Disclosure has to be "clear and distinguishable" and given "at the latest at the time of the first interaction or exposure." A label buried three scrolls down doesn't count. And if your AI talks to people, like a chatbot, you have to tell them they're dealing with a machine unless that's already obvious. (EU AI Act, Article 50)

One detail to watch. The May 2026 AI Omnibus agreement gave generative systems already on the market before 2 August 2026 until 2 December 2026 to meet the machine-readable marking duty. (Sidley, June 2026) The Commission is finalising guidelines and a Code of Practice on how to mark and label content in practice. (European Commission)

Does this reach you outside the EU? Yes, if your content reaches people in the EU. The Act follows the audience, not the head office.

What does the FTC require in the US?

The US has no single AI-labelling statute. What it has is sharper: a rule that makes fake and AI-generated reviews illegal outright.

The FTC's final Rule on the Use of Consumer Reviews and Testimonials took effect on 21 October 2024. It bans reviews and testimonials that misrepresent the reviewer as someone who doesn't exist, which directly covers AI-generated fake reviews, or who never used the product. It also bans buying, selling, or distributing such reviews when you knew or should have known they were fake. (FTC, August 2024)

The penalty is what makes this real. The FTC can pursue civil penalties of up to $53,088 per violation (the 2025 inflation-adjusted figure, raised most years), and each day of an ongoing violation can count separately. (FTC, civil penalty adjustments 2025) In December 2025 the FTC warned ten companies about possible breaches of this rule, so enforcement isn't hypothetical. (FTC, December 2025)

The wider FTC principle is older and broader: an endorsement has to reflect a real, honest experience. An AI-generated testimonial dressed up as a real customer fails that test whether or not anyone has written a dedicated AI rule.

Do UK advertising rules force AI disclosure?

This is where the original version of this article, and a lot of vendor marketing, gets it wrong. There's no blanket UK law that says "label your AI ads."

The ASA and CAP confirmed in 2025 that the existing CAP and BCAP Codes apply to all ad content, whatever made it. AI doesn't create a new disclosure rule. The old rules about being truthful, not misleading, and being clearly identifiable as advertising still apply in full. (ASA/CAP guidance)

So the test isn't "did you use AI." It's two questions CAP tells advertisers to ask:

1. Would the audience be misled if you didn't say AI was used?
2. If so, does disclosing AI clarify the message, or contradict it?

A blunt warning from the regulator: disclosure won't rescue a fundamentally misleading ad. You can't make a false claim and then disclaim it with "AI was used." That fails anyway. (ASA/CAP guidance)

My view: the UK position is actually harder to game than a fixed rule, because it turns on whether real people were deceived, not on whether you ticked a box.

What do the platforms require, on top of the law?

Platform rules are separate from the law and often stricter. They apply even where no statute does, and breaking them costs you reach, monetisation, or your account.

Platform	What needs a label	What happens if you don't
YouTube	Realistic altered or synthetic content (AI voice clones, deepfakes, altered real events). Effective 21 May 2025	Policy strikes, possible demonetisation (YouTube Blog)
TikTok	Realistic AI images, audio, or video; substantial AI edits to what a person says or does	Immediate strikes for unlabelled AI content; auto-detects via C2PA Content Credentials (TikTok Newsroom)
Meta (Facebook/Instagram)	Photorealistic AI images; auto-labels content made with Meta AI tools and reads C2PA metadata	Reduced distribution, labels applied for you

TikTok adopted C2PA Content Credentials in January 2025 and has since labelled over a billion AI videos using embedded metadata and watermarking. (TikTok Newsroom) That points at where all of this is heading.

How does C2PA fit in?

C2PA (the Coalition for Content Provenance and Authenticity) is the technical standard that makes machine-readable disclosure work. It's how a label survives a re-upload instead of getting stripped out.

Content Credentials, built on C2PA, attach signed, tamper-evident provenance data to a file: what made it, when, and what was edited. The v2.1 specification, published in 2024, hardened the binding between the credential and the content, so removing the credential breaks the signature. (C2PA Specification 2.1)

Why this matters for compliance: the EU AI Act asks for marking that's "machine-readable" and "interoperable." C2PA is the leading way to meet that without inventing your own scheme. Adopt it now and you're building toward the August 2026 standard rather than scrambling for it.

Building a disclosure process that holds up

Four moves cover most of the exposure across these regimes:

• Tag at creation. Mark every asset as AI-generated, human-made, or hybrid the moment it's produced. You can't disclose what you didn't track.
• Match the strictest rule that applies. For a given asset, that's usually the EU AI Act (if you reach EU audiences) or the platform you're posting to. Meet the highest bar and the lower ones are covered.
• Embed, don't just caption. Use C2PA Content Credentials so the disclosure travels with the file and survives re-uploads.
• Keep the receipts. Hold records of what was AI-made, where it was published, and how it was labelled. That documentation is your defence if a regulator or platform asks.

For teams using AI image tools specifically, the asset-level governance gaps are covered in our guide to AI image generation compliance risks. For the broader marketing picture, see AI marketing compliance risks.

Frequently asked questions

When does AI content disclosure become mandatory in the EU?

The core transparency duties in Article 50 of the EU AI Act apply from 2 August 2026. Generative systems already on the market before that date have until 2 December 2026 to meet the machine-readable marking duty, under the May 2026 AI Omnibus agreement. (EU AI Act, Article 50)

Do I have to label every AI-assisted edit?

No. The EU AI Act exempts assistive editing that doesn't substantially change the input. Platforms take a similar line: TikTok and YouTube target realistic content that could mislead, not minor touch-ups like colour correction or cropping. (YouTube Blog)

Is there a UK law requiring AI disclosure in ads?

Not a blanket one. The ASA and CAP confirmed the existing advertising codes apply to all content regardless of how it was made. You disclose AI when not doing so would mislead the audience, judged case by case. (ASA/CAP guidance)

What's the penalty for an undisclosed AI-generated review in the US?

The FTC's rule treats AI-generated fake reviews as illegal, with civil penalties of up to $53,088 per violation under the 2025 inflation-adjusted figure, and each day of an ongoing breach can count as a separate violation. (FTC, August 2024)

The bottom line

Treat disclosure as a workflow step, not a legal afterthought. The smart move is to build to the EU AI Act's machine-readable standard now, using C2PA, because it's the strictest bar and it satisfies the platforms at the same time. The teams retro-fitting labels in late 2026 will pay more and move slower than the teams that tagged content at source all year.

One honest caveat: these rules are still settling. The Commission's Article 50 guidelines and Code of Practice aren't final, and platform policies shift often. Build a process you can adjust, not a one-off fix. And get the call right where the stakes are highest, deepfakes, regulated sectors, and anything that could mislead a buyer, because that's where the penalties and the reputational hit land hardest.

VerityAI advises boards and marketing leaders on responsible AI use, including synthetic content governance that keeps you compliant without killing the speed AI gives you. If your team is publishing AI content across jurisdictions, we'll help you build the disclosure process before the deadline does it for you.

More in VerityAI's disclosing AI-generated content in gaming.

If you want support with this, VerityAI offers our AI governance practice.