The EU AI Act makes you label deep-fake ads, not every AI ad

Since 2 August 2026, the EU AI Act has required the advertiser or agency that uses AI to make an ad to disclose when its image, video or audio is a deep fake: content that resembles real or realistic people, objects, places or events and "would falsely appear to a person to be authentic or truthful". It does not require a label on every AI-generated ad. The European Commission's guidelines count an AI product image that makes the product look better than it is as a deep fake, and a real car on an AI-generated background as not one. Breaking these transparency rules can cost up to €15 million or 3% of worldwide annual turnover, whichever is higher.

That is the half of the answer our post on Meta's rules left open. Meta doesn't make you disclose AI in an ordinary commercial ad. The EU can, and its rule is law, not platform policy. It binds the advertiser or agency that made the ad, not the platform that shows it.

I read the regulation, the Commission's guidelines on Article 50 in full, its FAQ and the code of practice that goes with them. Every claim below is quoted, with the article or paragraph number.

This is not legal advice. It reports what the documents say. The guidelines describe themselves as "non-binding" and note that "any authoritative interpretation of the AI Act may ultimately only be given by the Court of Justice of the European Union". Several of their answers are, in their own words, "case-by-case".

Do I have to label AI-generated ads in the EU?

If they are deep fakes, yes. Otherwise generally not; AI-written copy is the one edge case, covered below. Article 50(4) of the AI Act puts the duty on the "deployer":

Deployers of an AI system that generates or manipulates image, audio or video content constituting a deep fake, shall disclose that the content has been artificially generated or manipulated.

Article 50 has no equivalent sentence for AI images, video or audio that aren't deep fakes.

The obligation people confuse with it belongs to the AI tool, not to you. Article 50(2) requires providers of generative AI systems to make sure their output is "marked in a machine-readable format and detectable as artificially generated or manipulated". That is a mark software reads, not a label a viewer sees. The Commission's guidelines say deployers "cannot rely on the machine-readable marking embedded in the content by the provider" to meet their own duty (paragraph 117).

Article 50 also covers chatbots (a duty on the provider) and emotion recognition and biometric categorisation systems (a duty to tell the people exposed to them). Neither is a label on ad creative.

Who is the deployer: the brand, the agency or the platform?

Article 3(4) defines a deployer as anyone "using an AI system under its authority except where the AI system is used in the course of a personal non-professional activity". The guidelines use advertising as their worked example (paragraph 14):

Where the deployer of an AI system is a legal person under whose authority the system is used (e.g. an advertising company), the individual employees that act under the instructions and under the control of that legal person (e.g. digital animators, web designers, content creators, journalists) should not be considered as separate deployers for that system.

The company stays the deployer when it hands the work to "contractors, freelancers" under its control. A brand that hires an agency is a different case:

By contrast, a company that merely commissions an advertising agency to produce an advertisement, without taking decisions and exercising control over whether and how the advertising agency uses AI in the production process, is not a deployer.

So the answer depends on who decides whether and how AI is used. If the brand decides, the brand is the deployer. If the agency decides, the agency is.

Platforms are not deployers when their role "is limited to disseminating or transmitting AI-generated or manipulated content created by third parties" (paragraph 16).

Being outside the EU doesn't take you out of scope. The deployer duty reaches companies in third countries that foresee their output being used in the EU, "by directing or authorising distribution within the Union, including by posting deep fakes on the globally accessible internet" (paragraph 13). The guidelines' own example: "a company established in a third country that uses an AI system to generate a deep fake of a celebrity featured in an advertisement displayed in the Union is also a deployer".

What counts as a deep fake in an ad?

Article 3(60) defines a deep fake as "AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful". The guidelines read that as four cumulative criteria (paragraph 113). The Commission's FAQ, which spells it "deepfake", compresses them to three.

1. Resemblance. "A high level of similarity" to the subject. "The content does not need to be identical to the subject."
2. Existing. Nothing has to be real. "It is sufficient for simulated persons, objects, places, entities or events to resemble someone or something that exists, can plausibly exist or could have plausibly existed in reality." What falls outside is the unrealistic: "humans flying without mechanical aids, dragons, or elephants driving cars".
3. Persons, objects, places, entities or events. Persons include "realistic AI-generated human avatars or personas" and "personal characteristics or expressions, such as image, voice". Objects include "consumer goods". Events include "the depiction of professional or consumer services".
4. False appearance of being authentic or truthful. Authenticity covers "the involvement of real human beings", "the actual appearance or use of objects or the delivery of services". Truthfulness "pertains to the veracity of the content".

Three details in paragraphs 114 and 115 matter for ads. The test is objective and "does not require the intention of the deployer to deceive". Photorealism makes a deep fake more likely, "but photorealism alone is not determinative". And the judgment isn't made for an average viewer: if the content could deceive part of the "reasonably foreseeable audience", such as children or the elderly, that "may suffice".

The advertising examples in the guidelines

The product-image rows are the ones to read twice. In both, the guidelines' test is whether the ad could mislead about the product, not whether AI made it.

Is AI editing of a real product photo a deep fake?

Usually not, if the product itself isn't changed. Paragraph 116 says AI edits to "insignificant substantive or technical aspects of pre-existing content" may not make it a deep fake, and gives a sentence written for advertisers:

For example, AI-powered colour correction, background extensions of existing content, adjustments or replacements of backgrounds for clearly aesthetic purposes, compositions and arrangements of existing products, or re-scaling of images applied in product advertisements or packaging is likely to have only a minor impact on a person's perception of the authenticity and truthfulness of the advertisement and the product.

"Likely" is the guidelines' word, and the same paragraph says the answer "depends on the context and the impact on the persons' perception".

There is a second editing exemption, and it is often confused with this one. Article 50(2) says the provider's marking duty "shall not apply to the extent the AI systems perform an assistive function for standard editing or do not substantially alter the input data provided by the deployer or the semantics thereof". The guidelines define standard editing as "preparing existing content for publication or distribution" that "does not involve generating new content" (paragraph 90), and list examples on both sides:

That table is about the tool's machine-readable mark under Article 50(2). Whether your finished ad is a deep fake that you must label is the separate test above.

Can an ad count as creative work and get a lighter label?

Sometimes, but the guidelines read the exception strictly. Article 50(4) limits the duty for deep fakes that form part of "an evidently artistic, creative, satirical, fictional or analogous work or programme" to disclosure "in an appropriate manner that does not hamper the display or enjoyment of the work". It is still a disclosure. The lighter regime changes how you disclose, not whether.

Paragraph 122 reads "evidently" strictly. It excludes content whose nature "is exclusively informative or commercial and is recognisable as such", and on ads it says:

Some kinds of content (e.g. advertisements or documentaries) containing deep fakes might be regarded as evidently creative or fictional works in certain, specific situations, but not in others, since the assessment is case-specific. When the deep fake combines multiple characters (e.g. informative and creative), the informative character should always prevail and require compliance with the standard labelling requirements.

Both advertising examples in the guidelines' list of deep fakes that are not creative works are product demonstrations: the teleshopping video and the synthetic influencer testing a sponsored product.

What does the disclosure have to look like?

The Act sets the standard, not the design. Article 50(5) says the information must be given "in a clear and distinguishable manner at the latest at the time of the first interaction or exposure" and must meet accessibility requirements.

The guidelines add three points that bear on ads:

• Perceivable without effort. Labels should be "understandable and perceivable by natural persons (e.g. with visible or audible labels), without them needing to rely on any specific technical tools or performing dedicated actions" (paragraph 117).
• Not buried. Information is not clear and distinguishable "where it can be easily overlooked or missed by natural persons under normal exposure or interaction conditions (e.g. only included as part of a manual or hidden under layers of menu options on an online interface, part of terms of use that are often not read by users)" (paragraph 142).
• Not only at the start. The duty "applies to each output of an AI system with respect to any natural person exposed to the content". And "if it is reasonably foreseeable that persons may not perceive content from its beginning, then only disclosure at the beginning of content does not adequately inform those persons and should be complemented with disclosure at later moments, where possible" (paragraph 143). One of its examples of first exposure is "when encountering AI-manipulated deep fakes when scrolling on social media".

The Code of Practice on Transparency of AI-generated Content is where a design appears. It is final, published on 10 June 2026, and the Commission (8 July) and the AI Board (9 July) assessed it as adequate. It is also voluntary. For deployers who sign it, it specifies an icon whose main element is "the capitalised acronym 'AI'", offers free EU icons reading "AI + GENERATED" and "AI + MODIFIED", and gives placements such as "the top right corner of an image or video deep fake". For video, the label appears "at the beginning of the video as well as, where possible, at regular intervals". For audio, the code calls for "a short audible disclaimer" at the start. Companies that don't sign, says the Commission's FAQ, "will have to demonstrate compliance through alternative adequate means".

Does Meta's AI info label count?

Neither document says. Here is what each one does say.

Meta's help page on AI info: "In most instances, you will be able to find the AI info label by navigating to the three-dot menu, selecting About this ad, and then tapping on the AI info label."

The guidelines (paragraph 142): information is not clear and distinguishable if "hidden under layers of menu options on an online interface".

The guidelines (paragraph 126), on very large online platforms: "Where providers of VLOPs or VLOSEs make labelling tools available to such deployers enabling them to label their deep fake content in compliance with Article 50(4) AI Act (i.e. a clear and distinguishable disclosure of the AI-origin), those deployers can rely on such tools". The paragraph adds that this is "without prejudice to the responsibility of the deployers".

Meta's page again: in regions whose laws or guidelines include AI transparency requirements, "advertisers will have the opportunity to self-disclose that GenAI was used and if they do so, the label will appear on the ad next to the Sponsored label". The listed regions include "the European Region".

Meta's page doesn't mention the AI Act or claim that either label meets it, and the guidelines don't mention Meta. How Meta detects AI in ads covers where the label appears.

Does AI-written ad copy need a label?

Generally not. The second subparagraph of Article 50(4) covers AI text "published with the purpose of informing the public on matters of public interest". It is exempt where the text "has undergone a process of human review or editorial control" and someone "holds editorial responsibility". The guidelines list among texts outside the rule:

AI-manipulated text that is part of a company's advertisement or product descriptions (not including any claims related to e.g. health, consumer safety or sustainability).

Read the brackets carefully. They take health, safety and sustainability claims out of the "outside" example. They do not say such copy is automatically in. That still turns on the three tests in paragraph 131: published text, meant to inform the public, on a matter of public interest. The guidelines' list of such matters includes "public health" and "environmental protection". And the human-review exemption has a floor: the guidelines treat fact-checking as the minimum (paragraph 134), and a spell-check or a cursory approval doesn't count (paragraph 135).

What about AI voices in audio ads?

Audio is covered: the deep-fake rule applies to "image, audio or video content", and "persons" includes a person's "voice". The guidelines' audio examples sit on both sides:

• Deep fake: "AI-generated audio involving voice cloning of a newspaper podcast's regular presenters and a guest interlocutor discussing some news events."
• Not a deep fake: "AI voice replication for fictional characters (e.g., for audiobooks, games or animation) when there is no deception as to the identity of the narrators."
• Not a deep fake: a radio broadcast where AI only adjusts "technical audio parameters (e.g. normalising volume levels, noise reduction, audio compression) without altering the actual words spoken".

The guidelines give no example of a synthetic voiceover in a radio or streaming ad. The four criteria above are the test they leave you with. Where the code of practice applies, an audio deep fake carries "a short audible disclaimer in plain and simple natural language" at the start, plus a visual label "when a screen is available".

When did this start, and is there a grace period?

Article 113 of the Act: "It shall apply from 2 August 2026." Article 50 is not among the parts given a different date.

The only transition is for providers. The Digital Omnibus on AI, Regulation (EU) 2026/1744, published in the Official Journal on 24 July 2026, added Article 111(4):

Providers of AI systems, including general-purpose AI systems, generating synthetic audio, image, video or text content, that have been placed on the market before 2 August 2026 shall take the necessary steps in order to comply with Article 50(2) by 2 December 2026.

That covers machine-readable marking only. The deployer's deep-fake disclosure has applied since 2 August 2026, with no transition.

Deep fakes made before that date don't have to be labeled retroactively. Paragraph 154 says deep fakes "generated or manipulated before 2 August 2026 do not need to be marked or labelled retroactively". Text is treated differently: if text generated before that date is published on or after it, "they need to be labelled". Labeling older material is encouraged, "without being expected to engage in disproportionate efforts to that end (such as auditing of pre-existing content databases or modifying already printed product packaging)".

What are the fines?

Article 99(4)(g) covers "transparency obligations for providers and deployers pursuant to Article 50": up to "EUR 15 000 000 or, if the offender is an undertaking, up to 3 % of its total worldwide annual turnover for the preceding financial year, whichever is higher". For SMEs, including start-ups, it is whichever is lower (Article 99(6)). The Omnibus extends that to small mid-caps. National market surveillance authorities do most of the enforcing, according to the Commission's FAQ.

A label also doesn't make an ad lawful. The guidelines say the disclosure duty "does not imply that AI-generated or manipulated deep fakes that are harmful and unlawful under the applicable Union or national law (e.g. misleading advertising ...) may be generated and disseminated" (paragraph 129). Recital 137 of the Act says the same about compliance in general.

Where the summaries go too far

The overstatement to watch for is "everything AI-made needs a label now". One advertiser guide, published on 15 October 2026, says of AI images, video and audio that those made before 2 August 2026 "don't need a label now", then adds: "Anything generated since does." (The same guide gets the car example right elsewhere.)

The Act doesn't support the general version. Article 50(4)'s label duty covers deep fakes. The guidelines list AI-generated ad content that isn't a deep fake, the talking mice and the real car on an AI background, so by their own examples not everything AI-made needs a deployer label. Most AI output does have to be machine-readably marked, with the editing exceptions above. But that is the tool provider's duty, for systems already on the market it runs from 2 December 2026, and it is not a label a viewer sees.

How this fits with Meta's rules

Two systems are asking two different questions. Meta asks whether AI made or edited the ad, and for commercial ads it applies its own label, with nothing you are required to declare. The EU asks whether the ad is a deep fake that could pass for real, and if it is, the disclosure duty is yours, whatever the platform does.

Sources

• Regulation (EU) 2024/1689 (Artificial Intelligence Act) — EUR-Lex, Official Journal text of 12 July 2024: Article 3(3), 3(4) and 3(60) definitions, Article 50 in full, Article 99(4)(g) and 99(6) penalties, Article 113 application dates, recitals 132–137 (rendered in a browser, accessed 2026-10-16)
• Regulation (EU) 2026/1744 (Digital Omnibus on AI) — EUR-Lex, Official Journal of 24 July 2026: the new Article 111(4) transition to 2 December 2026 for providers' Article 50(2) marking, and the SMC fine rule; confirmed it does not amend Article 50(2), 50(4) or the deep-fake definition (rendered in a browser, accessed 2026-10-16)
• Guidelines on transparency obligations for providers and deployers of AI systems — European Commission, C(2026) 5054 final, 20 July 2026: every paragraph number quoted above. The Communication approving them says the Commission approved their content on 20 July, that formal adoption comes "at a later date, when all language versions are available", and that "it is only from that moment that these Guidelines will be applicable"; the library page says the Commission "adopted" them (both PDFs from the Commission's library page, read in full, accessed 2026-10-16)
• Transparency obligations under Article 50 of the AI Act — European Commission FAQ, last updated 24 July 2026: the three-criteria summary of a deep fake, non-signatories of the code, enforcement by national authorities, no retroactive labeling (fetched, accessed 2026-10-16)
• Code of Practice on Transparency of AI-generated Content — European Commission: the final code (published 10 June 2026), Section 2 for deployers: the "AI" icon, the EU icons, placement and audio disclaimer (page fetched, code PDF read in full, accessed 2026-10-16)
• Commission opinion on the assessment of the Code of Practice on Transparency of AI-generated Content — European Commission, 9 July 2026: the Commission's adequacy conclusion of 8 July and the AI Board's assessment the following day (fetched, accessed 2026-10-16)
• About AI info on ads created or edited with generative AI tools — Meta Business Help Center: the three-dot-menu location of the AI info label and the self-disclosure option for the European Region (rendered in a browser, accessed 2026-10-16)

---

The regulation and the Omnibus were read on EUR-Lex in a real browser, which returns nothing to a plain fetch. The Commission's guidelines (about 50 pages) and code of practice are PDFs linked from its library pages and were read in full, not from summaries. Where the guidelines and the FAQ differ in wording, this post quotes the guidelines and notes the difference. The guidelines say the Commission will review them "as soon as necessary", so check the current version before relying on any of it.

Related: what Meta actually requires you to disclose and how Meta detects AI in ads, and what the label costs you.

Terrads makes image, video and audio ads for Meta, TikTok and other placements, with your brand library carried into every one.