The European Union's AI Act reached a significant enforcement milestone on August 2, 2026, when the European Commission's AI Office gained the power to investigate and fine providers of general-purpose AI models, alongside enforcement of the law's bans on prohibited AI practices and its transparency requirements for certain systems.
The shift moves the AI Act from a law defined mostly by future compliance dates into one with active investigative and penalty powers, even though other major provisions — including the bulk of the rules governing "high-risk" AI systems — are not scheduled to become enforceable until later, with some pushed to December 2027.
Why it matters
Frontier AI labs including OpenAI, Google DeepMind, Anthropic and Meta now face direct exposure to EU enforcement action over how their general-purpose models are documented, tested and disclosed, rather than facing only the prospect of future obligations. Because GPAI-model fines can reach 3% of a company's worldwide annual turnover, the number carries real weight even for the largest AI developers, and the AI Office's choices about early enforcement priorities will shape how seriously the broader industry treats the law's later phases.
How the rule works
The AI Act, formally Regulation (EU) 2024/1689, phases in obligations over several years. Provisions banning prohibited AI practices, requiring certain transparency disclosures, and governing general-purpose AI models under Articles 53 through 55 became enforceable on August 2, 2026, giving the AI Office and national authorities the ability to investigate potential violations and impose penalties. Rules covering high-risk AI systems used in areas like employment, credit and law enforcement follow a separate, later timeline.
Evidence
The European Commission's own AI Act Service Desk confirms that "some enforcement powers of the AI Office and the authorities of the Member States will start to apply on 2 August 2026," specifically naming prohibited practices, transparency rules and GPAI provisions, while independent legal trackers such as Taylor Wessing and Bratby Law separately confirmed the same effective date and fine ceiling in analyses published around the enforcement start.
The competing read
EU officials describe the phased rollout as a deliberate, risk-based approach that gives industry time to prepare for the toughest obligations while still holding the most powerful AI systems to account early. Some AI companies and trade groups have argued the compliance burden, even in this early phase, adds cost and legal uncertainty in Europe relative to less-regulated markets, while some digital-rights advocates counter that delaying the high-risk system rules to 2027 leaves the law's most consequential protections dormant for too long.
What happens next
The AI Office is expected to signal its early enforcement priorities through initial investigations or guidance in the coming months, and companies are watching closely for the first GPAI-related fine, which would set an important precedent for how aggressively Brussels intends to use its new authority. Additional AI Act provisions, including rules for high-risk systems and further prohibited-practice categories, are set to phase in through 2027.