In May 2024, the EU Council formally approved the Artificial Intelligence Act (AI Act), marking the implementation of the world's first comprehensive AI regulatory framework. The Act adopts a risk-based classification approach, dividing AI applications into four tiers: unacceptable risk, high risk, limited risk, and minimal risk.
Applications categorized as "unacceptable risk" (such as social scoring systems and real-time remote biometric identification) will be completely banned. "High-risk" AI systems (such as recruitment screening, credit assessment, medical diagnosis assistance) must meet strict transparency, data governance, and human oversight requirements. General-purpose AI models (such as GPT-4, Claude) must comply with copyright law and training data transparency rules.
For AI startups and tool developers, the Act brings compliance costs but also creates new market opportunities — AI auditing, explainability tools, and compliance consulting will become rapidly growing new sectors. Violations could result in fines up to 7% of global annual revenue.
The Act will be implemented in phases over the next 6-24 months, giving enterprises different compliance transition periods. The US and China are also expected to introduce their respective AI regulatory frameworks around 2025.