What the EU AI Act means for AI content production

The EU AI Act, formally the European Artificial Intelligence Regulation, lays down common rules for developing and using AI systems across the European Union. For anyone producing content with artificial intelligence, the provision that matters most is Article 50, which has applied since 2 August 2026.

That article introduces transparency obligations for text, images, video and audio generated or manipulated with AI.

What does the AI Act require? Content labelling obligations#

The Regulation separates two distinct obligations.

The first is technical marking. Providers of generative AI systems have to make their outputs identifiable as artificially generated or manipulated, in machine-readable form. In practice that can mean metadata, watermarks, technical markers or provenance standards such as C2PA and Content Credentials.

The second is visible or audible labelling. This obligation sits with the companies, public bodies or professionals who publish AI-generated content.

Not every AI-assisted piece needs a visible label. The obligation is aimed mainly at content that could be considered a deepfake.

Which images, videos and audio need to be labelled?#

Content can count as a deepfake where it realistically depicts an existing or plausible person, object, place, entity or event, in a way that could lead an audience to believe it’s genuine.

Official criteria for classifying content as a deepfake#

Cases that should generally be labelled include:

• A synthetic image of a real person doing something that never happened.
• A photograph manipulated to add, remove or replace people or objects in a way that changes its meaning.
• A video using face swapping or lip syncing to make someone appear to say words they never spoke.
• An AI recreation of a war, an accident, a demonstration or an event, presented as genuine footage.
• A cloned voice used to attribute false statements to an identifiable person.
• An artificially generated interview, call or conversation presented as authentic.

Exceptions: technical retouching, minor edits and fantasy works#

By contrast, clearly fantastical illustrations, abstract imagery, unmistakably fictional characters, and technical corrections that don’t materially change the meaning of a piece, shouldn’t automatically be treated as deepfakes.

Adjusting colour, removing noise, upscaling resolution or normalising volume can fall outside the obligation, provided none of it alters the message or creates a false impression of authenticity.

Where does that leave advertising and artistic work?#

Artistic, satirical and fictional works aren’t fully exempt. Where they incorporate a deepfake, transparency still applies, though the disclosure can be worked in without spoiling the experience of the piece.

In advertising, the test is whether the audience could reasonably read the piece as an authentic depiction of a person, product, place or event. Where something is published matters just as much as how it was made.

For the procedure itself, see the guide on documenting and certifying AI-generated content step by step, and the official EU icons for the visible side of it.

How does VeriqX support Article 50 compliance?#

VeriqX documents how a piece of AI content was produced. The platform records the tools and models used, the prompts, the input files, the modifications made, and the degree of human involvement.

From that evidence it issues a certificate bound to the final file, using eIDAS-compliant timestamping and provenance standards such as C2PA / Content Credentials.

VeriqX doesn’t determine whether content is true, lawful or good — that’s not its job. Its job is to provide the traceability that shows how a piece was made, what controls were applied, and why it was labelled, or wasn’t.

Getting ready for the AI Act isn’t a matter of adding a label at the end. It means building transparency, documentation and control into the creative process from the start.

Key points and frequently asked questions about the EU AI Act and AI content#

When did content labelling become mandatory under the EU AI Act?#

On 2 August 2026, when the Article 50 transparency requirements for synthetic content became binding.

What is the difference between technical marking and visible labelling?#

Technical marking is machine-readable metadata, such as C2PA. Visible or audible labelling is a direct disclosure to the viewer that the content is a deepfake.

How does VeriqX help you apply the AI Act and label AI-generated content?#

VeriqX documents the creative process along with the tools and inputs used, and issues a certificate with eIDAS timestamping that evidences the diligence a company applied.

Need to bring your audiovisual or advertising production in line with the EU AI Act?#

• Technical consultancy and corporate certification: [email protected]
• Talk to the author directly (David Lahoz): https://www.linkedin.com/in/dlahoz/