EU Gains Powers to Regulate Major AI Models
The European Union will gain sweeping new powers effective August 2 to police large general purpose AI models. From that date, the Commission will have the authority to demand data from model providers, carry out safety assessments, mandate corrective actions, remove models from the EU market, and impose penalties reaching up to 3% of annual total turnover.
This move places the EU at the forefront of global AI safety regulation. However, the effectiveness of these powers hinges on the Commission’s willingness and ability to enforce them. Strong enforcement would signal Brussels’ commitment to upholding AI rules, even if it creates friction with the United States or China, where most leading model providers are based.
The AI Act enters its enforcement era
The launch of these new powers on August 2 has been years in the making. Throughout 2026, the EU worked to simplify its digital rulebook, including the EU AI Act. In May, lawmakers agreed to delay rules governing high-risk AI systems and remove industrial applications from the scope, but left the timetable for general purpose AI models untouched.
The rules target providers of the largest general purpose AI models, with a threshold so high that only a handful of companies, such as OpenAI, Anthropic, Meta, Alphabet, and xAI from the U.S., and Alibaba, ByteDance, and Z.ai from China, are expected to fall within scope. France’s Mistral is the only European model provider affected.
The AI office has already built much of the infrastructure needed to enforce the rules, including publishing a Code of Practice, issuing guidelines, and establishing an AI Act Advisory Forum. However, concerns remain about its capacity, with estimates suggesting it may need up to 160 staff by 2030, compared to around 40 as of July.
Read Also: Meloni marks four years as Italian leader
Initial enforcements by the AI office are unlikely to involve major fines. Instead, they will focus on requests for information and structured dialogue with model providers. However, Brussels ultimately has the power to suspend providers from the EU market and issue fines worth hundreds of millions.
Companies that refused to sign the Code of Practice, such as Meta and xAI, may face closer scrutiny. The AI office is also likely to focus on models with the highest impact capabilities, such as those from OpenAI and Anthropic, which have recently sparked concerns about cyber capabilities and loss of human control.
Success for the AI office will depend on its ability to prioritize the largest safety risks, adapt to rapid technological changes, and avoid politicization. With international support growing for AI regulation, the EU is not alone in this endeavor, as countries like the U.S. and China also move towards greater oversight of frontier models.
International Context and Future Outlook
The EU’s move towards AI regulation is not isolated. Other countries are also taking steps to oversee frontier models. In the U.S., states like Illinois, New York, and California are developing their own AI rules, while China has expressed its intention to lead in global AI governance.
The AI Office’s Enforcement Challenges and Strategies
The AI Office faces significant challenges in enforcing the AI Act, including limited personnel and the need to scrutinize some of the world’s most valuable companies. As of July, the unit had around 40 staff, far below the estimated 160 needed by 2030. This shortage raises concerns about its capacity to handle enforcement effectively. The office’s initial actions are expected to focus on information requests, structured dialogue with model providers, and independent evaluations rather than immediate fines.
