Since August 2, the European Commission has gained real leverage over developers of general-purpose artificial intelligence (GPAI) systems. This refers to the full activation of the penalty mechanism for violations of the Artificial Intelligence Act (AI Act), which has now moved from the declarative stage to the practical realm. This is a key moment: a year of obligations, and only now can the regulator enforce compliance.
What changes for the market
Obligations for GPAI model providers have been in effect since August 2025, but only now is the European Commission authorized to enforce compliance through financial sanctions. Companies are required to maintain technical documentation, disclose data to downstream providers, and strictly adhere to EU copyright norms. For the most powerful models carrying systemic risks, there is an additional requirement to notify the European AI Office and conduct vulnerability assessments. New players entering the market after August 2, 2025, already fall under these rules; for others, the deadline is August 2, 2027.
Size of fines: from 1% to 7% of turnover
The penalty structure is differentiated. For violating GPAI obligations, refusing to provide documents, or submitting incorrect data—up to €15 million or 3% of global annual turnover (Article 101). A stricter regime under Article 99 provides for up to €35 million or 7% of turnover for prohibited practices, including social scoring and manipulative systems. Intermediate violations, for example in the field of high-risk systems, are punishable by up to €15 million or 3%, while misleading regulators incurs up to €7.5 million or 1%. For SMEs, a lower threshold applies, which is logical but still significant.
Transparency and content labeling
The new requirements oblige developers to clearly label deepfakes and any synthetic content, as well as to warn users about interactions with AI. This also applies to emotion recognition systems. For generative models released before August 2, 2026, there is a transition period until December. The voluntary Code of Practice, signed by Anthropic, Google, Microsoft, OpenAI, and others, remains a tool for demonstrating compliance but does not replace the law. Notably, xAI signed only the safety chapter—the rest will have to be proven manually.
High-risk systems—pause until 2027
Rules for high-risk AI are postponed until December 2027, and for AI embedded in medical devices and toys—until August 2028. This is a deliberate step following the Digital Omnibus, giving businesses time to prepare standards. The regulator has also launched tools for filing complaints and for whistleblower operations.
My comment: Europe is deliberately playing the long game, giving giants time to adapt while creating a precedent for global regulation. The question is whether fines can keep pace with AI development, or whether this is another bureaucratic compromise that the market will bypass around the corner.