Since August 2, the European Commission has gained real enforcement powers in the field of artificial intelligence. The regulator is now entitled to impose fines on providers of general-purpose AI models (GPAI) for violations of the Artificial Intelligence Act (AI Act). This is a landmark moment: a transition from declarative norms to practical enforcement.
Key changes and timelines
Obligations for GPAI model providers have formally been in effect since August 2025, but only now has the European Commission gained the authority to enforce them. Companies are required to document the technical parameters of their systems, disclose data to downstream providers, and comply with European copyright law. For the most powerful models posing systemic risks, a separate requirement is provided — notifying the European AI Office and conducting risk assessments.
Importantly: for providers that entered the market after August 2, 2025, the rules are already mandatory. Those who appear later are granted a deferral until August 2, 2027.
Penalties: two regimes
The penalty system is divided into two levels. Under Article 101 of the AI Act, violations of GPAI provider obligations carry a fine of up to €15 million or 3% of global annual turnover. This applies to refusal to provide documents, submission of incorrect data, or failure to implement requested measures.
A stricter regime is Article 99. For prohibited practices, including social scoring and manipulative systems, the fine reaches €35 million or 7% of turnover. Other violations, such as non-compliance with requirements for high-risk systems, are punishable by up to €15 million or 3%. For providing false information to national regulators — up to €7.5 million or 1%.
Transparency and labeling
The new requirements oblige developers to clearly inform users about interaction with AI, including emotion recognition systems. AI-generated or AI-modified content must receive machine-readable labels. For generative systems released before August 2, 2026, a transition period applies until December.
Code of Practice and signatories
The voluntary Code of Practice remains an important tool for demonstrating compliance. Among the signatories are Anthropic, Google, Microsoft, OpenAI, and other giants. Notably, xAI agreed only to the safety chapter, which leaves questions about its position on transparency.
What's next
Rules for high-risk systems will only take effect in December 2027, and for AI in physical products — in 2028. The European Commission has also launched tools for filing complaints and working with whistleblowers.
My analysis: The delay in enforcement for high-risk systems is a deliberate compromise, giving businesses time to adapt. However, the market should prepare for the EU to actively use its new powers specifically in the GPAI segment, where clear guidelines already exist. Companies, especially those relying on voluntary codes, should reconsider their compliance strategies before the end of the year.