AI Labelling Obligations: What Has Applied Since 2 August 2026

Since 2 August 2026 the transparency obligations of the EU AI Act apply. Whether an AI-generated item must be labelled depends on the role a company takes on, the type of content and the exemptions in the regulation. Agencies and brands should therefore clarify who deploys the AI system and which obligations follow from that.

What has applied since 2 August 2026?

The EU AI Act (Regulation (EU) 2024/1689) sets out information and labelling obligations for certain AI systems and their content in Article 50. The obligations are staggered: the prohibitions on certain AI practices have applied since February 2025. Article 50 has applied since 2 August 2026. For machine-readable marking by providers of AI systems already placed on the market before that date, the European Commission allows a transition period until 2 December 2026.

The regulation distinguishes two roles: providers develop AI systems or have them developed and place them on the market under their own name. Deployers use an AI system in the course of their business. Both carry different obligations.

What has to be labelled?

The following overview shows four use cases of Article 50 that are relevant to marketing and communications. Further obligations exist, for example for emotion recognition and biometric categorisation.

Chatbot

People must be informed that they are interacting directly with an AI system, unless the use of AI is obvious in the context (Art. 50(1)). Obligation of the provider.

AI-generated media

Audio, image, video and text must be marked in machine-readable form (Art. 50(2)). Obligation of the provider. The obligation does not apply where the system merely supports standard edits and does not substantially alter the content of the input data.

Deepfake

Must be disclosed (Art. 50(4)). Obligation of the deployer. Deepfakes are AI-generated or AI-manipulated image, audio or video content that resembles persons, objects, places or events and falsely appears authentic or truthful. For obviously artistic, satirical or fictional works, disclosure may be made in a way that does not spoil the work.

AI text on public affairs

Must be disclosed (Art. 50(4)). This covers texts published to inform the public on matters of public interest. The obligation does not apply if a natural or legal person has reviewed the content and assumes editorial responsibility.

Timing

The notice must be clearly recognisable no later than the first interaction or first contact with the content and must meet accessibility requirements (Art. 50(5)).

What does this mean for advertising and AI video?

Article 50 regulates two different things: the technical marking by the provider and the disclosure of certain content by the deployer. For advertising this means: a notice such as "This scene was generated with AI" must clearly identify the passage concerned and be perceptible at first contact. For a silent clip a text overlay is enough; for an audio ad a voice-over or a notice in the description text.

Provenance metadata can be lost on export and upload. The team should therefore check the published file on each channel. For the disclosure of deepfakes: the use of AI must be perceptible to the audience. A purely machine-readable marking does not fulfil this obligation.

Examples of machine-readable markings: C2PA documents the origin and editing steps of a piece of content; SynthID embeds watermarks in the content. Suitable verification tools can read such data. A missing marker does not rule out AI generation, however.

What happens in the event of violations?

Article 99(4)(g) makes violations of the transparency obligations subject to fines. For companies the ceiling is 15 million euros or 3 per cent of worldwide total turnover in the preceding financial year, whichever is higher. For small and medium-sized enterprises the lower of the two ceilings applies (Art. 99(6)). These are maximum limits, not automatic fines.

The regulation is enforced by the national market surveillance authorities. In Germany the Federal Network Agency (Bundesnetzagentur) is fundamentally responsible, as the AI Market Surveillance and Implementation Act (KI-MIG) stipulates. The law provides for other supervisory authorities for certain sectors; for advertising, media-law requirements and competition-law claims can also be relevant. Companies should document the review of their obligations, the tools used and the labels actually published.

What we do at QLTY

We produce video, music and text with AI. For every project we deliver documentation of the AI tools used and the intended labels. In an initial meeting we record the AI content used and discuss which labelling questions must be reviewed before publication.

FAQ

Does AI music in advertising have to be labelled?

Two aspects must be assessed separately: the machine-readable marking, which the provider of the generator must apply, and disclosure in the advertising medium. A disclosure obligation under Article 50(4) exists for audio content classified as a deepfake. Whether an AI-generated song falls under this depends on whether it falsely makes a performance or recording appear authentic, for example.

Does this also apply to existing campaigns?

According to the European Commission, content created before 2 August 2026 does not have to be labelled retroactively. New AI content within an existing campaign must be assessed separately. Document the creation date of every motif and any later AI edits.

Is a notice in the legal notice enough?

No. The information must be recognisable at first contact with the content. A general notice in the legal notice (Impressum) is not sufficient if the audience does not receive the required information when viewing or listening to the content.

Who monitors compliance in Germany?

Fundamentally the Federal Network Agency (Section 2 KI-MIG). The law provides for other supervisory authorities for certain sectors, for example for media services. For advertising, media-law requirements and competition-law claims can additionally be relevant.

The legal information is based on Regulation (EU) 2024/1689 (Articles 50 and 99), guidance from the European Commission and the German KI-MIG. This article is not legal advice. The full text is available on EUR-Lex.