Starting August 2, the European Commission has officially gained the authority to impose sanctions on providers of general-purpose AI models (GPAI) for violations of the Artificial Intelligence Act (AI Act). This is a landmark step that turns previously declared norms into a real enforcement tool.
Key changes in enforcement
Obligations for providers of GPAI models have been in effect since August 2025, but only now can the regulator enforce compliance through financial penalties. Companies are required to document the technical parameters of their systems, disclose data to downstream providers, and comply with copyright rules. For the most powerful models with systemic risks, they must notify the European AI Office and implement threat assessment measures.
It is important to note: for providers that entered the market after August 2, 2025, the rules are already mandatory. For others, the deadline is August 2, 2027.
Fine amounts: two levels of liability
Under Article 101 of the AI Act, which specifically concerns GPAI models, the maximum penalty will be up to 15 million euros or 3% of global annual turnover. This applies to refusal to provide documentation, submission of false data, or failure to comply with risk assessment requirements.
However, Article 99 provides for stricter measures—up to 35 million euros or 7% of turnover—for prohibited practices such as social scoring or manipulative systems. For providing incomplete information to national regulators, the penalty is up to 7.5 million euros or 1% of turnover. A lower threshold applies to small and medium-sized businesses.
Transparency and labeling: new requirements
Transparency rules also come into force from this date. Chatbots must inform users about interaction with AI, and deepfakes and synthetic content must receive machine-readable labels. For generative systems placed on the market before August 2, 2026, a transition period until December is provided.
Separately, I note the voluntary Code of Practice signed by Anthropic, Google, Microsoft, OpenAI, and others. xAI joined only the safety chapter, which will force it to demonstrate compliance with transparency and copyright norms through other means.
What's next
Rules for high-risk systems will only come into force on December 2, 2027, and for AI in physical products—in August 2028. The European Commission has also launched tools for filing complaints and for whistleblowers.
My conclusion: the EU is moving from a phase of "soft regulation" to real enforcement. This will set a precedent for the global market—companies will have to reconsider their compliance strategies, especially those working with powerful models. However, the deferral for high-risk systems shows that the regulator is aware of the complexity of preparing infrastructure, and this is a reasonable balance between innovation and control.