As advances in AI make it increasingly difficult to distinguish between genuine and fabricated content, the question is no longer what AI is capable of creating, but whether people can recognize what is real

Transparency in AI Use Transparency is a core principle of Article 50 of the EU Artificial Intelligence Act (AI Act). Rather than restricting AI systems, the provision requires disclosure so users know when they are interacting with AI-generated or AI-manipulated content. AI systems that interact with humans must clearly identify themselves. AI-generated content, including text, images, audio, and video, must be labeled to prevent confusion with human-created material. Deepfakes and AI-generated content related to matters of public interest must also disclose their artificial origin. In addition, individuals must be informed when AI is used for emotion recognition or biometric categorization, as these technologies can affect how people are assessed and treated. More than a technical requirement, Article 50 aims to reduce the risk of deception. When users cannot distinguish between human and AI-generated content, their ability to make informed judgments may be undermined. Transparency therefore helps preserve trust and maintain a clear boundary between humans and machines in an increasingly AI-driven world. Yet this raises a practical question: how can these transparency requirements be translated into something that people can actually see, recognize, and understand in their everyday digital lives? From principle to practice In May 2026, the European Commission (EC) released draft guidelines on implementing the EU AI Act, highlighting the practical realities of AI transparency. While Article 50 establishes the principle, the guidelines tackle a more concrete question: how can AI-generated content be effectively labeled and identified in the real world? The answer is not straightforward. Watermarks on AI-generated images can be removed, while metadata identifying a file’s origin can be altered or stripped away. To address these limitations, the EC proposes a layered approach that combines visible labels, digital fingerprints, and secure tagging mechanisms to improve content traceability. In the AI era, transparency is no longer a simple disclosure requirement but a technical infrastructure designed to support trust and accountability. Addressing deepfakes Rather than imposing an outright ban, the EU AI Act requires AI-generated content to disclose its artificial origin, while exempting works created for artistic, satirical, or parody purposes. This reflects a policy choice to address the risks of deepfakes without unduly restricting creative expression or social criticism. Yet the exemption creates a challenge. The more realistic satirical content becomes, the more effective it is, but also the more likely it is to be taken out of context and mistaken for reality. As content is shared across digital platforms, contextual cues may disappear while the impression of authenticity remains. Therefore, the balance between risk control and the protection of freedom of expression is not a stable endpoint, but rather a constantly shifting state, dependent on how content is created, disseminated, and received within the digital environment. This very balance also raises a practical question: do users actually pay attention to transparency labels? Or will they become another form of “notice fatigue”, similar to cookie notices, where users routinely click “accept” without reading the content? At a deeper level, Article 50 of the EU AI Act and its accompanying guidance are not only about transparency, but also about trust. In a world where content can be generated at scale at virtually no cost, trust no longer stems from the content itself, but from the ability to trace its origin. In other words, the crucial question is no longer “Is this content true or false?”, but rather: “Where does it come from?” Transparency and labeling In recent years, deepfake videos, synthetic voices, and “generative” advertising content have begun to appear on social media, sometimes being used to attract viewers and sometimes to facilitate fraud. Meanwhile, the current legal framework still focuses primarily on unlawful content after it has already occurred, rather than designing transparency mechanisms from the outset. This reveals a policy gap: as content becomes increasingly difficult to verify with the naked eye, relying solely on moderation or enforcement may no longer be sufficient. From this perspective, the EU’s approach suggests a different path forward: rather than attempting to control content itself, the focus may need to shift toward designing transparency and provenance mechanisms at the system level. For businesses, particularly digital platforms and AI developers, the message is also quite clear: transparency is no longer merely an ethical choice, but is increasingly becoming an obligation that can be verified through technology. As AI becomes increasingly human-like, requiring it to “identify itself” may be the minimum step needed to preserve something fundamental: the human ability to distinguish. Yet the question remains: in a world where everything can be labeled, will we still take the time to read those labels? (*) PhD in Law, Lecturer at University of Economics and Law, VNU-HCM The views expressed in this article are those of the author and do not necessarily reflect the views of the institution with which the author is affiliated.