Fake doctors offering wrong advice. Finance ministers suggesting investments. Faces and voices of celebrities asking for help for someone. Fake videos are increasingly numerous, from India to America, from Russia to Europe, also because they are easily made with generative artificial intelligence. The epidemic of photos of public figures portrayed in indecent clothes caused by the lack of modesty of the control systems of Elon Musk’s model, Grok, filled social media at the beginning of 2026.
Read more Ft: US funds to European groups aligned with Maga movement
Ask the Sun
The questions are automatically suggested by 24Ore AI
based on the content viewed.
For Europe, it is time to take countermeasures. Starting from August 2, the rules provided by the AI Act, article 50, come into force, which oblige companies that produce artificial intelligence systems and those who use them to create and publish texts, photos, videos, audio, to take responsibility. And now, thanks to the work of a team of independent experts, the Commission also explains how to reduce the resulting risks.
In practice, the general principle is that producers of generative artificial intelligences and those who use them to produce content to be published must be transparent and ensure that users know they are interacting with the synthetic result of artificial intelligences. For system producers, there is an obligation to mark generated content with a sort of digital watermark or other solutions that are readable by special software and allow recognizing the artificial origin, for example, of images or videos. For those who had systems available on the market before August 2, the obligation probably shifts to December 2. Those who produce and publish content made with artificial intelligence, in turn, from August 2, must warn the public, except for exceptions provided by law, using special labels.
Read more Organic label for products made with Italian raw materials
But it is not always technically easy to comply with these obligations. Those who produce content must warn the public if they use artificial intelligence but without risking the enjoyment of the work. And those who make the systems must mark the products of their artificial intelligence, but in a way that is simple to use and recognize, possibly with standard and interoperable tools. To facilitate the task of the parties involved, the Commission has appointed a group of experts who have produced a code of practices, that is, a sort of manual that allows those who follow it to demonstrate that they have done everything possible to avoid ambiguous interpretations of content generated with artificial intelligence.
Today the code is presented in preview. Dino Pedreschi, a professor at the University of Pisa, was among the leaders of the expert group. “The code of practice is part of a complex attempt to push towards a less polluted information ecosystem. If things work, a good part of the synthetic material will be transparent. And in the long run, it will also be easier to prove the human origin of the content.” It is about evolving the use of metadata that qualify the information circulating on the web. “Our code of practice is accepted and applied voluntarily,” explains Pedreschi, “but those who adopt it are sure to know what they must do to comply with this part of the AI Act.” Consequently, among other things, it will be easier if one has to defend oneself in court for issues related to this kind of matter. Cases will be judged in the courts to which any injured parties have turned. In some cases, the Data Protection Authority is also competent. The code of practice will be a help for producers and publishers who voluntarily adopt it and who find themselves having to defend themselves. OpenAI has already signed. Meta, for now, has refused.”
Read more Tremonti: «Italy-China Commission established, room for parliamentary diplomacy»
