The European AI Act Imposes Strict Restrictions on High-Risk Systems

The European AI Act (EU AI Act) came into full effect in early 2025, becoming the world's first comprehensive legislative framework for artificial intelligence and establishing a system that classifies AI applications according to four risk levels: acceptable, limited, high, and prohibited.

Completely prohibited applications include comprehensive social scoring systems for citizens, emotion recognition in workplace and educational environments, and subliminal manipulation of human behavior, while healthcare systems, autonomous vehicles, and critical infrastructure are subject to strict prior audit requirements.

Major tech companies such as Google, Meta, and Microsoft face a deadline to comply with transparency requirements for large language models, or face fines of up to 35 million euros or 7% of their annual global revenue, whichever is higher.

EU officials believe the law will be exported globally just as the GDPR privacy law was, pointing out that companies wishing to access the European market will automatically comply with it, creating a de facto global standard for regulating AI even outside Europe's borders.

What do these terms mean?

Risk classification in the EU AI Act: The law divides AI applications into four categories: acceptable such as games, limited risk such as chatbots, high risk such as healthcare, and prohibited such as social scoring.

GDPR: The European privacy law enacted in 2018 became a de facto global standard for data protection, as companies around the world complied with it to serve European customers.

Large Language Models (LLMs): AI models trained on massive amounts of text that excel at generating content and answering questions, most notably ChatGPT, Gemini, and Claude.