Chatbots, deepfakes, AI: Europe imposes new rules

Since August 2, 2026, Article 50 of the AI Act has entered into force. The European Commission’s text pursues a clear objective: to enable internet users to better distinguish AI-generated or AI-modified content from authentic content. But can these new rules actually be applied at a time when artificial intelligence tools are ubiquitous?

Content generated or modified by AI must be labeled as such

Providers of AI systems must develop a labeling system to identify content generated or modified using artificial intelligence, whether it is a video, an image, sound, or text. Deepfakes disseminated in a professional context must also be labeled as such. The objective is simple: to enable internet users to distinguish content generated or modified by artificial intelligence from real content.

Furthermore, chatbots used by certain platforms, such as public services or e-commerce websites, must be clearly identified as artificial intelligence systems. As a result, an internet user must be able to quickly understand whether they are interacting with AI or a human. It should be said that some platforms maintain ambiguity in this regard by passing AI off as a genuine customer service representative.

Another particularly important rule: texts of public interest generated by AI must be labeled as such. This point concerns only texts published without human editorial oversight. A person who uses AI to write a text but who rereads and approves its content is exempt from this rule. Artistic, satirical, or fictional texts are also exempt from this rule.

Will the labeling of AI-generated content really be effective?

Several platforms have already introduced a labeling system to identify content generated or modified with the help of AI. For example, Instagram offers “AI notices.” Meanwhile, the AI Act requires AI-generated content to “be marked in a machine-readable format and identifiable as having been generated or manipulated by AI,” but current technologies (digital watermarks, metadata, provenance systems) have their limitations. Digital watermarks can, for example, be removed, while metadata associated with files can disappear during editing or reposting.

In addition, where is the line between content substantially modified by AI and a simple touch-up? AI tools are now ubiquitous and have been widely made available to the general public. In the long term, the risk is therefore of seeing a proliferation of “modified with the help of AI” labels on the web.The European Commission and the platforms concerned will have to clarify certain practical aspects to ensure the consistent application of these new rules.

👉🏻 Follow the latest tech news in real time: add 01net to your sources on Google and subscribe to our WhatsApp channel.

Source: European Commission