Falsità sulla trasparenza: cosa offrono e cosa non offrono le nuove etichette AI

The European Union’s new AI regulations will soon take effect, introducing mandatory labels for content generated by artificial intelligence. On Sunday, parts of the AI Act, designed to increase transparency and combat misinformation, will come into force. Under these rules, providers must clearly label synthetic content such as text, images, videos, or audio. The goal is to help users distinguish between human-generated material and AI-created content, thereby fostering public trust and reducing the spread of false information. The regulations stem from the EU's Artificial Intelligence Act, which has been gradually implemented since its adoption over two years ago. Final guidelines on who is required to comply with the labeling rules and how specific types of content should be marked were recently published. While private individuals using AI tools to create personal items like holiday cards or emails are exempt, businesses and professionals utilizing AI models face more stringent obligations. These include both model developers and organizations deploying AI systems for commercial purposes. Despite the recent clarification, many industry experts argue that the rules still contain ambiguities. For instance, the guidelines define scenarios where AI-generated content does not require labeling. This includes situations where AI is used merely as a supportive tool, such as grammar checking or video format conversion, as long as the original material remains largely unchanged. However, the boundaries between human work and AI output continue to blur, particularly in creative fields like music production. Some companies, including German music industry players, have already announced plans to introduce logos identifying AI-assisted or fully AI-generated musical works. While some see this as a helpful step toward consumer clarity, others warn that the increasing integration of AI in artistic processes makes it increasingly difficult to determine the origin of a given piece. Critics point out that the current framework fails to address certain cases where the artificial nature of content is immediately apparent. For example, a promotional video featuring animated mice discussing cheese would not need explicit labeling, as its non-human origin is evident. Conversely, manipulated images showing real footballers standing in a stadium where they never played would be classified as deepfakes and thus require clear marking. Such distinctions highlight the challenge of defining what constitutes misleading or deceptive AI-generated content under the new rules. Raphael Fischer, a researcher at the Technical University of Dortmund, acknowledges that mandatory labeling is a “useful building block” for promoting transparency in online content. He emphasizes, however, that trust in AI systems depends not just on visible labels, but also on broader factors such as the openness and accountability of AI technologies themselves. According to Fischer, meaningful trust requires not only transparency in content labeling but also transparency in how AI systems operate and how they are managed responsibly. As the AI Act moves forward, the effectiveness of these measures will depend on how well they are enforced and adapted to evolving technological practices. With AI becoming more deeply embedded in everyday life, the challenge lies in balancing regulation with innovation while ensuring that users can reliably discern the origins of the digital content they encounter.