64 The Act came into effect on August 1, 2024 and will be fully enforceable in August 2026. Certain provisions (mostly prohibition on use) took on February 2, 2025. On February 4 and 6, 2025, the European Commission (EC) published guidelines on prohibited AI practices (here) and guidelines on AI system definition (here). At the same time, the EC withdrew its proposed AI liability directive (February 11, 2025). In this article we highlight the key prohibited practices under Article 5 of the AI Act, identify critical risks, offer actionable compliance insights and highlight enforcement mechanisms to help businesses remain ahead in this ever-evolving regulatory landscape. We also provide a quick reference to US developments as a counterpoint to the EU. Understanding the Scope of Prohibited AI Practices The AI Act defines four different levels of risk: 1. Unacceptable risk. Practices involving AI systems that are deemed to pose unacceptable risks on the basis that they are incompatible with fundamental rights and EU values are prohibited. 2. High risk. AI systems posing high risks to health, safety and fundamental rights can be placed on the market, put into service or used, subject to fulfilling certain requirements and obligations. 3. Limited risk. AI systems which are subject to transparency requirements (i.e., AI systems which perform autonomous tasks, are intended to interact directly with an individual or create content viewed by an individual, but which do not qualify as “high risk”). 4. Minimal risk. AI systems posing minimal to no risk are not regulated. Providers can adhere to voluntary codes of conduct The guidelines published by the EC on February 4, 2025, focus on the first category – practices involving unacceptable risk. Separate guidelines are expected for the second and third categories. Prohibited Practices The AI Act prohibits “manipulative, exploitative, social control or surveillance” AI-enabled practices, which by nature “violate fundamental rights and Union values.” Eight types of practices are identified, and sometimes illustrated by examples:
RkJQdWJsaXNoZXIy MjgzNzA=