EU Adopts Landmark AI Act
The European Union has officially adopted its landmark AI Act, establishing a new risk-based regulatory framework for artificial intelligence. The legislation introduces compliance reporting, transparency, and user rights, such as the ability to contest automated decisions. Product development for AI systems will now need to incorporate these compliance requirements from the design and roadmap stages, rather than as a post-launch consideration.
The world's first major law regulating AI, the Act has an extraterritorial reach, affecting any company offering AI-based products or services in the European Union, regardless of its location. This legislation is expected to influence global conversations and approaches to AI governance, setting a precedent for other jurisdictions. Penalties for non-compliance are substantial, with fines for violations involving prohibited AI practices reaching up to €35 million or 7% of a company's total worldwide annual turnover, whichever is higher. This exceeds the penalties established under GDPR. Lesser violations, such as providing incorrect information to authorities, can result in fines of up to €7.5 million or 1% of global turnover. The regulation categorizes AI systems into four tiers based on their potential risk: unacceptable, high, limited, and minimal. Unacceptable-risk applications, such as government-run social scoring and real-time biometric surveillance in public spaces, are banned. High-risk systems, including those used in healthcare or critical infrastructure, face stringent requirements like mandatory conformity assessments and robust documentation. The Act also introduces specific regulations for general-purpose AI (GPAI) models, like GPT-4 and Gemini Ultra. It establishes a two-tiered approach, with stricter obligations for "high-impact" models that pose systemic risks due to their power and influence. These requirements include assessing and mitigating potential risks. The implementation of the AI Act is phased. The ban on prohibited systems took effect in early 2025. Rules governing general-purpose AI models will apply by August 2025, with the majority of the Act's provisions becoming fully enforceable by August 2026.
