Artificial intelligence regulation (almost) adopted

The AI Act, or European regulation on artificial intelligence, was the subject of a political agreement between institutions on Friday night. Technology companies say they are disappointed by a text which, they say, will prevent certain innovations from being born and/or used in Europe, while the cultural sector is delighted to have obtained an obligation for transparency of LLM training data. « The political agreement on the AI Regulation took a long time for two reasons: firstly because new provisions on general-purpose AI (generative AI, Editor’s note) were introduced by the Parliament, and also because it took time to agree on the use of AI systems by law enforcement, » a European Commission official explained today at a press briefing. What emerges from last week’s trialogue, which ended on Friday night, is a text classifying tools using AI by risk: unacceptable risk (risks to fundamental rights, such as predictive policing AI), high risk (in the medical field and critical infrastructure, for example), specific risk linked to transparency (conversational robots, or deepfakes in particular, which will have to mention the use of AI), and minimal risk. The Commission states that « the vast majority of AI systems fall into the category of minimal risk systems. » It specifies that these systems are, for example, AI-based recommendation systems or spam filters, which « will benefit from a free pass and a lack of obligations, as these systems present minimal…