Today marks the first working day on which additional transparency obligations under the AI Act fully apply. Compliance is therefore no longer solely a matter for AI developers and providers, who face substantial fines for non-compliance. The new rules also affect ordinary users of artificial intelligence in the course of their professional activities, ranging from lawyers, marketing agencies, and HR departments to public media organizations.

The AI Act has not entered into force all at once but has instead been implemented gradually in several stages since February 2025. Nevertheless, a number of myths and misconceptions continue to surround its requirements. Some of the most common include the following:

Myth: An architect uses AI to create a visualization of a future office building. The image must be labelled “AI GENERATED”.

Reality: Typically not. The obligation does not apply to all AI-generated content, but rather to AI deepfakes and certain AI-generated texts published in the public interest. The key question is not whether AI was used to create the content, but whether the content could be perceived as an authentic record of reality. In this case, it is clear that the image depicts a proposed future building rather than an actual event or existing structure.

Myth: An architect uses AI to create a photograph of a completed office building that is intended to resemble a picture taken during the final inspection process, even though construction has not yet begun. No “AI GENERATED” or “AI MODIFIED” label is required because the purpose was not to mislead anyone but merely to present the project in an attractive way.

Reality: Intent is not decisive. The AI Act does not assess whether the creator intended to deceive anyone. If a photograph could reasonably be perceived as an authentic record of reality, despite depicting an event that never occurred, it may qualify as an AI deepfake and trigger a transparency obligation requiring appropriate labelling.

Author: Hana Císlerová