AI Content Labeling Rules Take Effect in the EU

Translated by a model; the original is in Russian.

· Artificial intelligence
As of August 2, companies are required to tell users when they are talking to artificial intelligence and to label AI-generated deepfakes. Violations carry fines of up to €15 million.

cover.svg

The European Union has introduced new transparency obligations under its dedicated artificial intelligence law — the AI Act. The rules took effect on August 2 and are meant to help people tell chatbots and AI-generated deepfakes apart from genuine content and a live interlocutor.

Who has to disclose what

The requirements differ for two categories of market players. Developers of AI systems (providers) must design them so that users are explicitly notified when they are interacting with artificial intelligence rather than a human — “unless this is obvious.” In addition, synthetic audio, images, video and text must carry embedded machine-readable markers that make it possible to recognize generated or altered material.

The platforms and services that use such systems (deployers) are required to label any AI-generated or AI-manipulated content — images, audio and video deepfakes that look real. Some companies, such as Meta and SpaceXAI, fall into both categories at once.

According to the European Commission, the rapid development of generative and interactive AI systems is increasingly blurring the line between the machine-made and the human. People need to understand when they are talking to an AI or encountering its output — this helps them make informed decisions, build a reasonable degree of trust and avoid falling for disinformation.

Ready-made icons and fines

So that platforms don't have to invent their own labels, the Commission has developed a set of ready-made marker icons and provided examples of how to use them. These icons echo the labels TikTok, Instagram and Facebook already have. Using the European icons specifically is not mandatory, but the Commission pointedly reminds everyone that the labeling requirements themselves most certainly are.

For failing to comply with the new rules, companies face fines of up to €15 million (about $17.2 million) or 3% of annual worldwide turnover. For new AI systems the rules apply immediately, while models and services launched before August 2 are given a transition period until December 2.


Source: The Verge

Comments

Sign in, to comment.

  • No comments yet.