From 2 August 2026, the transparency requirements set out in Article 50 of the EU AI Act will apply.
For those working in communications, this is no minor matter. Generated images, videos, synthetic voices, text and chatbots are now tools routinely used by agencies, companies and in-house marketing departments.
The question, therefore, is no longer simply whether to use artificial intelligence, but how to do so whilst maintaining a clear distinction between authentic content, generated content and modified content.
Why this regulation exists
Generative artificial intelligence now makes it possible to create extremely realistic images, videos, voices and text.
This offers enormous potential from a creative and productive perspective. However, it also raises an obvious problem: it is becoming increasingly difficult for those receiving content to tell whether what they see or hear actually happened or has been artificially generated.
The AI Act addresses precisely this issue by introducing specific transparency requirements.
This does not mean that every piece of content created using AI must be accompanied by a label.
Rather, it means that, in certain cases specified by the legislation, anyone who comes into contact with such content or an AI system must be able to recognise it as such.
It is not just about images
Article 50 distinguishes between different situations and liabilities.
Images, audio and video – deepfakes
When content generated or manipulated by AI realistically depicts existing people, objects, places, entities or events and may appear authentic, it falls within the scope of the regulations governing deepfakes.
Consider, for example, the voice of a director being used to utter words they never spoke, an altered photograph of a real person, or an artificial representation of an event presented as having actually taken place.
In such cases, the content must be clearly identifiable as having been generated or manipulated by artificial intelligence.
For works that are clearly artistic, creative, satirical or fictional, the legislation provides for less intrusive disclosure requirements, precisely so as not to compromise the enjoyment of the work.
Texts
For texts, the requirement is more limited. It applies to content generated or manipulated by AI and published with the aim of informing the public on matters of public interest.
However, the legislation provides for an important exception where the content has undergone human review or editorial oversight and a person or organisation assumes responsibility for its publication.
Advertising copy, a product description, an article or a post – even if developed with the support of AI – but which have been verified and approved by an editor, do not therefore automatically become content that must be labelled.
Chatbots and virtual assistants
Interaction with an AI system must also be transparent.
When a person communicates directly with an artificial intelligence system, they must be informed that they are interacting with a machine, unless this is already evident from the context.
Providers and users: two different sets of responsibilities
Another important aspect of the AI Act is the distinction between those who make artificial intelligence systems available and those who use them.
Providers of generative systems must implement technical solutions to ensure that generated or manipulated content can be identified as such, including through machine-readable labels.
Those who use these tools professionally and publish certain types of content, on the other hand, have specific disclosure obligations in the cases provided for by the legislation, such as deepfakes and certain publications of public interest.
This is an important distinction because, in day-to-day practice, content production is increasingly decentralised.
Some content is created in-house at the agency. Some is produced directly by clients.
Some may be generated by platforms, software or tools used internally by various company departments.
The management of AI therefore inevitably becomes a shared responsibility between the agency and the client.
What does this mean, in practical terms, for a business?
In our view, the point is not simply to apply an ‘AI’ label to any content. That would be an oversimplification and, in many cases, a mistake. Instead, a systematic approach is needed.
Understanding how content is created.
We need to remain aware of which assets have been generated or substantially modified using AI systems, and which are instead the result of standard processing, assistance or optimisation activities.
Distinguishing between instrumental and substantial use.
Correcting a translation, optimising an image or using AI to assist with writing is not the same as artificially creating a person, a voice or an event and presenting them as real.
Establish clear responsibilities.
Who produces the content? Who reviews it? Who approves its publication? And, where necessary, who decides how to disclose its origin?
These are questions that the agency and the client should address before publication, not once the content is already online.
How to flag AI-generated content
Where disclosure is required, the information must be clear and visible from the moment the content is first viewed.
The European Commission has also developed a system of icons specifically for content generated or modified using AI.
Their use is optional, but they provide a useful framework for developing, over time, a common and instantly recognisable visual language.
For IDP, it is above all a matter of trust
Artificial intelligence will play an increasingly prominent role in creative processes, marketing and communication. It would make little sense to seek to limit its use.
The real challenge lies in using it responsibly, whilst ensuring transparency regarding the distinction between what is created artificially and what is presented as real.
For an agency like IDP, this principle is not merely a new regulatory requirement. It concerns something that comes before technology: the credibility of communication.
Because whilst the tools may change, a brand’s task remains the same: to build, over time, a relationship of trust between what it says and what it actually is.
—
Legislative sources:
[Regolamento UE 2024/1689 (EU AI Act)]
[European Commission, EU Icons for labelling AI-generated content]
Did you enjoy our articles?
Subscribe to the newsletter!
