New Labels for AI: But critical thinking is needed

A voice on the phone, a chat agent, a video of a controversial speech by a well-known public figure. Telling the difference between what is human and what is artificial is an exercise in intuition that is becoming increasingly difficult but, at the same time, increasingly necessary. Without this ability, the polarization of public opinion is fueled, affecting above all the most vulnerable people, who lack the protections and tools needed to understand and respond to technological innovations.

Now, the difference will finally have to be made clearer. Since this month, some transparency obligations under the AI Act, the European regulation on artificial intelligence, have become applicable. Chatbots and virtual agents must immediately identify themselves as artificial intelligence systems when interacting with citizens. Similarly, synthetically generated or modified images, audio, and video must be clearly labeled and recognizable. Existing systems, on the other hand, will have to be adapted by December 2, 2026.

What has been done is a necessary step, but it is not enough to eliminate the risk of manipulation. A label will tell us that an image is artificial, but it does not explain why it was produced, who is spreading it, or whether its message is reliable. Transparency helps calibrate trust, but it does not replace the responsibility of those who publish content, nor does it replace our critical thinking: it is one piece of the puzzle, not a guarantee of truth, and it cannot become an excuse for shifting responsibility onto the end user.

This step, although important, also reveals another side of European AI policy: while the impact of artificial intelligence on people’s rights is already a reality, the strongest legal protections have been postponed. In fact, the application of the rules envisaged for systems considered high-risk to people’s rights (for example, those that may affect access to employment, education, or essential services, or that are incorporated into regulated products such as medical devices) has been postponed.

This postponement was motivated by the need to give companies and public administrations time to organize and adapt to technical standards that are still being developed. But the resulting message is clear and troubling: Europe is moving at two different speeds, putting on hold the most stringent safeguards designed to protect fundamental rights while AI spreads at a rapid pace through companies, offices, and society.

It is in this regulatory vacuum that training ceases to be a pursuit for specialists and becomes a tool for defending democracy. If the law postpones controls, the only immediate defense is critical literacy (AI literacy). A true culture of skills is needed to develop informed citizens and professionals capable of governing innovation. In this direction, FBK’s AI Compliance mission works to support the development of reliable artificial intelligence, together with important initiatives such as the first Summer School “AI + DATA: Practice & Law” organized in June by FBK with the OECD Centre in Trento, which attracted interest from participants from companies, public administrations, universities, and international organizations. At the center is not only an understanding of the rules, but also the development of a common language.

Therefore, AI will, in many cases, have to identify itself. But formal transparency is a bureaucratic illusion if it is not accompanied by political and social responsibility. And the defense of our freedoms does not come from a label; it depends on the ability of organizations to understand the tools they develop and use, and on the ability of citizens to question them critically.

Now we know when we are interacting with an artificial intelligence system; the next question we need to ask as a society is: who chose it, for what purpose, and with what checks and controls?