AI-generated content: what the new European regulation means for agencies

Since 2 August 2026, the transparency obligations of Article 50 of the European regulation on artificial intelligence (AI Act) have come into force. For communication agencies, content creators and advertisers using generative AI tools, this text concretely changes the way certain content is produced and distributed.

Four specific situations, not all AI content

Contrary to a common belief, Article 50 does not impose systematic labeling of all content created with the help of artificial intelligence. The text targets four well-defined situations: interaction with a chatbot, the technical marking of synthetic content (image, audio, video, text), the disclosure of deepfakes, and the information given during an emotion recognition or biometric categorization system.

Outside these four cases, no mandatory notice is required by the European regulation. An agency that uses AI to write an internal brief or generate a draft subsequently revised by a human therefore does not automatically fall within the scope of the obligation.

Two distinct obligations: marking and labeling

The regulation distinguishes two levels of responsibility:

• Marking, obligation of AI providers: making the content detectable as artificially generated in a machine-readable format (digital watermark, metadata).
• Labeling, obligation of deployers (agencies, publishers, brands): visibly informing users that a deepfake or a text related to a matter of public interest was generated or manipulated by AI.

This distinction is essential for web marketing agencies: the responsibility for labeling lies with the party that publishes the content to the public, not only with the publisher of the AI tool used to produce it.

A particular case: deepfakes and texts of public interest

The rules on deepfakes apply even in the absence of an intent to deceive: content that resembles a real person must be disclosed, even without any desire to mislead and even if it does not depict any existing person. Conversely, when the content is manifestly artistic, satirical or fictional, a minimal and non-intrusive disclosure is sufficient.

AI-generated texts about public interest topics, published without human review, are also subject to the labeling obligation. This point directly concerns newsrooms that automate part of their editorial production, an issue closely monitored by stakeholders in development and web marketing.

Sanctions that can be severe

Failure to comply with these transparency obligations exposes one to fines of up to €15 million or 3% of the company’s annual global turnover, whichever is higher. A level of sanction comparable to that provided for by the GDPR, which places AI compliance among the priorities for corporate legal and marketing departments, including SMEs in the B2B sector that use specialized providers.

For agencies, the issue is therefore no longer only editorial but also contractual: verifying that the AI tools used by their providers comply with marking obligations, and documenting the uses of content generated or manipulated by AI in their own deliverables.

Frequently asked questions

Is it necessary to disclose a simple text written with the help of a generative AI?

Not systematically. The labeling obligation only applies if the text concerns a matter of public interest and is published without human review. Content revised by a writer falls outside this case.

Who is responsible for labeling: the agency or the AI tool provider?

Both have distinct obligations: the tool provider must make the content detectable as AI-generated (technical marking), while the agency or brand that publishes the content must, in the relevant cases, clearly inform the public (labeling).

What are the sanctions in case of non-compliance?

Fines can reach €15 million or 3% of the company’s annual global turnover, a level comparable to the sanctions provided for by the GDPR.

Sources

• Leptidigital
• Fidealis
• EU Artificial Intelligence Act (official text, Article 50)

This article was written with the help of artificial intelligence. Editorial policy