AI Act and AI-generated video: what brands and agencies must do

Since 2 August 2026, the AI Act has required transparency for images, audio and video made with artificial intelligence. If you commission or produce AI-generated video, this guide covers what Article 50 actually says, who has to apply the label (brand or agency), when an AI video counts as a deepfake, how to label it, and a checklist for your next brief.

The picture is now complete: the regulation, the Commission’s guidelines, an approved code of practice and, in Italy, a new criminal offence for harmful deepfakes. None of these rules stop you using AI in an ad. They ask you to say so, and to have the right permissions.

What the AI Act requires of AI-generated video from August 2026

Article 50 of Regulation (EU) 2024/1689 splits the work in two. Providers of generative AI systems must mark their outputs in a machine-readable format. Deployers, meaning anyone using those systems in a professional capacity, must visibly disclose content that constitutes a deep fake, clearly and at the latest at first exposure. Under Article 113, all of this applies from 2 August 2026.

What the digital omnibus changed

Regulation (EU) 2026/1744, the Digital Omnibus on AI, in force since 27 July 2026, pushed the high-risk deadlines back to 2027 and 2028. It barely touched Article 50. Providers of generative systems already on the market before 2 August 2026 have until 2 December 2026 to implement technical marking. The duty to disclose deepfakes was not postponed, as the Commission’s Article 50 guidelines of 20 July 2026 confirm.

The code of practice and the guidelines

On 10 June 2026 the Commission published the Code of Practice on Transparency of AI-generated Content, which it assessed as adequate on 8 July. Signing is voluntary, and by the end of July it had around 190 signatories, including Google, Meta, OpenAI and the advertising group Dentsu. The guidelines are non-binding, but they explain the definitions with plenty of examples drawn from advertising.

Brand or agency: who has to label

The deployer is whoever decides whether and how an AI system is used, and takes responsibility for its outputs. The guidelines are explicit: an advertising company using AI is a deployer, even when it brings in freelancers. A company that merely commissions an ad, without deciding whether or how the agency uses AI, is not.

In practice the line moves easily. If the brand asks for a realistic avatar of its CEO, or chooses which scenes are generated, it is making exactly those decisions. The guidelines also expect deployers to make sure the label reaches the audience along the whole distribution chain, including through contractual terms. So who labels, where and in which versions belongs in the contract, not in a scramble once the campaign is live.

When an AI video is a deepfake

Under the AI Act, a deep fake is AI-generated or manipulated content that resembles existing persons, objects, places, entities or events and would falsely appear to be authentic. The guidelines extend “existing” to anything that could plausibly exist, so a photorealistic synthetic actor counts as a person. Photorealism alone does not settle it; context matters too.

Examples that need a label

• An ad featuring an AI-generated version of a well-known influencer.
• A realistic synthetic avatar of the CEO presenting the year’s results.
• An AI image that makes a product look better than it really is.

Examples that are not deepfakes

• Mice arguing about the best cheese, in a campaign for a cheese maker.
• A real car against a generated background, as long as the ad does not mislead about the product.
• Colour correction or background extension in product shots.

The creative exception is narrow

For deepfakes that form part of an evidently artistic, creative, satirical or fictional work, a disclosure that does not spoil the viewing is enough. It is not an exemption, and it is read strictly: an ad qualifies only in specific situations. According to the guidelines, a teleshopping-style video with synthetic people, or a realistic synthetic influencer trying out a sponsored product, is not a creative work.

How to label a video in practice

The technical marking embedded by the tool is not enough: viewers cannot see metadata or invisible watermarks. You need a visible or audible label. The code of practice proposes an EU icon built around the letters “AI”, including “AI + GENERATED” and “AI + MODIFIED” variants. For video, the guidance is specific:

• the icon embedded in the video, in a spot free of other graphics, for example the top right corner;
• visible at the start and, where possible, at regular intervals, and at least after every break;
• for audio-only content, such as a radio spot with a cloned voice, a short spoken disclaimer at the start;
• for creative works, a disclosure in the description, the credits or next to the player.

The regular intervals matter on social, where people drop into a video halfway through. If viewers may foreseeably miss the opening, a label at the start alone is not enough.

Platform labels

YouTube asks creators to disclose realistic altered or synthetic content in YouTube Studio, and may label files carrying C2PA metadata automatically. Meta applies its “AI info” label based on industry-standard signals or self-disclosure. TikTok requires creators to label realistic AI-generated content. The guidelines say you can rely on these tools if they produce a clear disclosure, but only within that platform. For your website, TV or out-of-home screens, the safest place for the icon is the video itself.

Italy: consent, likeness and the new deepfake offence

Italy’s Law no. 132 of 23 September 2025 added Article 612-quater to the Criminal Code: one to five years in prison for anyone who causes unjust harm to a person by disseminating, without their consent, images, video or voices altered with AI and capable of misleading people about their authenticity.

For advertisers the point is simple: a label does not replace consent, as both the guidelines and the code of practice stress. Under Italian copyright law, a person’s portrait cannot be reproduced or sold without their consent (Article 96), and data protection rules apply to anyone identifiable. Law 132 also clarifies that works made with the help of AI are protected only when they are the product of the author’s own intellectual work, so it pays to document the human creative contribution.

A checklist for commissioning AI video

1. Classify every video. Are there realistic people, products, places or events that could pass as real? Then it probably needs a label.
2. Decide who the deployer is. The contract states who applies the label, in which formats and on which channels.
3. Label the master. Icon at the start and at regular intervals, a spoken disclaimer for audio, plus each platform’s own disclosure.
4. Get consent before production. Written releases for faces and voices that cover AI use, duration, territories and channels.
5. No celebrity lookalikes. Without an agreement, a recognisable face or voice is a legal risk, label or no label.
6. Show the product as it is. AI can build the scene; it should not improve the product.
7. Keep the records. Tools, dates, consents, labelling decisions and delivered versions. Do not strip provenance metadata.
8. Train the team. Whoever publishes needs to know when a label is required and how to fix a missing one.

How we work at Kult

At Kult, rights and consent are agreed before the first frame. Digital casting is part of pre-production, and usage rights, revision rounds and delivery dates are written into the proposal before work starts. It is also the right moment to decide how the AI will be disclosed: a directing choice, not a patch at the end of the edit. If you are working with digital characters, start with our FAQ.

This article is general information, current as of October 2026, and is not legal advice.

Frequently asked questions

Do AI videos published before August 2026 need a label?

No. According to the Commission’s guidelines, deepfakes generated before 2 August 2026 do not need to be labelled retroactively, although labelling them is encouraged where it takes no disproportionate effort. Anything generated or manipulated with AI after that date is fully in scope.

What are the fines for not labelling a deepfake under the AI Act?

Breaches of Article 50 can lead to fines of up to EUR 15 million or 3% of worldwide annual turnover, whichever is higher. For SMEs and start-ups the lower of the two applies, and adherence to the code of practice can count as a mitigating factor.

Who enforces the AI Act’s transparency rules?

National market surveillance authorities enforce Article 50, alongside the EU AI Office for systems built on general-purpose models from the same provider. In Italy, Law 132/2025 designated the National Cybersecurity Agency (ACN) as market surveillance authority, and anyone can lodge a complaint.

Sources

1. Regulation (EU) 2024/1689 (Artificial Intelligence Act) · EUR-Lex, Publications Office of the EU, 2024-06-13
2. AI Act, Article 50: Transparency obligations for providers and deployers of certain AI systems · European Commission, AI Act Service Desk, 2026-07
3. AI Act, Articolo 50: Obblighi di trasparenza per i fornitori e i deployer di determinati sistemi di IA · European Commission, AI Act Service Desk, 2026-07
4. AI Act, Article 113: Entry into force and application · European Commission, AI Act Service Desk, 2026-07
5. AI Act, Article 99: Penalties · European Commission, AI Act Service Desk, 2026-07
6. Regulation (EU) 2026/1744 (Digital Omnibus on AI) · EUR-Lex, Publications Office of the EU, 2026-07-08
7. Guidelines on the implementation of the transparency obligations for certain AI systems under Article 50 of the AI Act, C(2026) 5054 · European Commission, 2026-07-20
8. Code of Practice on Transparency of AI-generated Content · European Commission, AI Office, 2026-06-10
9. Commission publishes Code of Practice on marking and labelling AI-generated content · European Commission, 2026-06-10
10. Code of Practice on Transparency of AI-generated Content: FAQ · European Commission, 2026-06-10
11. Commission opinion on the assessment of the Code of Practice on Transparency of AI-generated Content · European Commission, 2026-07-09
12. Strong backing for the Code of Practice on Transparency of AI-generated Content · European Commission, 2026-07-31
13. Legge 23 settembre 2025, n. 132 - Disposizioni e deleghe al Governo in materia di intelligenza artificiale · Gazzetta Ufficiale della Repubblica Italiana, 2025-09-25
14. Art. 96 Legge sul diritto d’autore (L. 22 aprile 1941, n. 633) · Brocardi.it, 1941-04-22
15. Disclosing use of GenAI content · YouTube Help, 2026-10
16. Our Approach to Labeling AI-Generated Content and Manipulated Media · Meta, 2024-04-05
17. Partnering with our industry to advance AI transparency and literacy · TikTok Newsroom, 2024-05-09