Not all content created by AI will be required to be labeled: What the new EU AI act requires

After an extremely long bureaucratic journey, finally, on August 2, most of the rules established in the European Union AI Act will begin to apply, aimed at regulating the use of this technology across our territory. Within this text, the European Commission has chosen to place the emphasis on the obligations related to ensuring transparency in its use, as set out in Article 50 of the law.

Through an official statement and various explanatory documents, Brussels is seeking to clarify several aspects that are worth knowing before it comes into force.

Not surprisingly, the Commission foresees fines of up to 15 million euros, or up to 3% of total worldwide annual turnover for large companies, and “proportionality” for SMEs and startups that fail to comply. That said, throughout the text you will see that the Commission regularly distinguishes between AI system providers (the companies that develop tools such as ChatGPT or Gemini) and those that use them in their professional activity. That distinction is essential for understanding which obligations apply to each party.

When will you have to disclose that someone is talking to an AI?

The first transparency obligation applies to systems that interact directly with people, such as chatbots, virtual assistants, or AI agents.

The general rule is simple: when a user is conversing with an artificial intelligence system, they must be informed clearly and from the outset, from the very first moment of the conversation or video call, unless that is obvious from the context. In other words, if an assistant looks like a robot and introduces itself as such, there is probably no need to overdo it. But if the conversation could be mistaken for one with a real person, disclosure will be mandatory.

The European Commission also stresses that this information must be adapted to the audience the system is aimed at, taking into account factors such as age or possible situations of vulnerability. In other words: it is not enough to hide a notice in the terms and conditions. The user must be able to easily understand that they are interacting with an AI.

What about images, videos, or audio created with AI?

This is the point that content creators, media outlets, and marketing departments at any company will be watching most closely.

System providers (the companies that develop ChatGPT, Claude, Gemini, and so on) capable of generating synthetic images, videos, or audio must ensure that this content includes a machine-readable technical marker that makes it possible to identify its artificial origin. The idea is that this information, this invisible watermark in the form of metadata, should travel with the file even when it circulates on the Internet.

The European Commission recommends that these markers be as resistant as possible to later modifications, although it acknowledges that there is no foolproof system and that the technology will continue to evolve. The goal is to facilitate the detection of AI-generated content without affecting the user experience.

At this point, as you can see, responsibility falls on the major developers and not on users who rely on tools such as ChatGPT, Gemini, or Claude to generate content.

Deepfakes will have to be clearly identified

One of the most controversial aspects of Article 50 concerns so-called deepfakes, for which the law provides a legal definition. It refers to “image, audio, or video content generated or manipulated by AI that resembles existing persons, objects, places, entities, or events and would falsely appear to a person to be authentic or truthful.”

In these cases, it is no longer enough to include a technical marker that may be invisible to the user. Whoever publishes this type of content must visibly disclose that it is material generated or manipulated using AI.

In other words, if you include a deepfake in one of your commercial initiatives, as part of your content strategy or marketing campaign, the responsibility to clarify that falls on you. The company or professional that uses and distributes the content must clearly and perceptibly disclose that it has been generated or manipulated using artificial intelligence. This information must be shown, at the latest, the first time the public is exposed to the content.

The Commission does clarify, however, that there are some exceptions. For example, when the use forms part of an artistic, satirical, or fictional work, the notice may be adapted so as not to undermine the audience’s experience, provided it remains sufficiently clear.

And what about AI-generated text?

All right, we are clear on images, video, and audio… but what about text?

The short version is that it is not as straightforward as we might have liked.

The European Commission makes it clear that not every text written with the help of artificial intelligence must be labeled. The obligation mainly applies to content intended to inform the public about matters of public interest when it has been generated or manipulated by AI and is published without significant editorial review by a person.

In other words, if a journalist, marketing professional, or any other author uses AI tools such as ChatGPT as support for drafting a text, buuuut then reviews it, corrects it, and assumes responsibility for the final result, that transparency obligation does not apply. In the Commission’s words:

• Superficial checks that are merely formal or procedural (for example, spell-checking or grammatical correction) are not considered human review or editorial oversight.
• Editorial responsibility means that a person must bear ultimate legal responsibility for the publication of the content, including human review or editorial oversight.

In this context, using ChatGPT, Gemini, Claude, or other tools as writing assistants does not automatically mean that every text must be labeled as AI-generated content. What matters is the degree of human intervention and who assumes editorial responsibility for the published content.

This rule opens the door to some debatable situations. For example, this obligation does not appear to generally affect product pages, technical specifications, or purely commercial ad copy. The rule focuses on publications intended to inform the public about matters of public interest. That said, exceptions could arise when the content concerns, for example, consumer safety or economic or health issues that are relevant to public debate.

As important as shopping may be… product descriptions, technical specifications in a catalog, or commercial ad copy fall outside the scope of that visible labeling requirement.

Or at least, that is my interpretation.

Let us recap

If you use AI in your company, these are the four questions you should be asking yourself starting August 2:

• Do my users know when they are talking to an AI?
• Do the tools I use to generate AI images, audio, or video include the required markers?
• Am I correctly identifying deepfakes when necessary?
• Do the contents I publish with the help of AI have sufficient human oversight for me to assume editorial responsibility for them?

Rules designed to build trust

The European Commission insists that these obligations are not intended to slow the development of artificial intelligence, but rather to increase the trust that citizens and businesses place in this technology. That is why the new rules focus on ensuring that people know when they are interacting with an AI and when the content they are seeing has been generated or manipulated using it.

Starting August 2, these obligations will begin to apply across the European Union. And although many companies were already using this type of notice voluntarily, transparency will stop being a best practice and become a legal requirement in many situations.

Image: ChatGPT