Since August 2, the European Commission has gained the authority to impose fines on providers of general-purpose artificial intelligence models (GPAI) for violations of the AI Act. This is a landmark step that transforms regulation from a declarative framework into a truly enforceable mechanism.
Simultaneously, new transparency requirements have come into force. Chatbots and interactive AI systems are now required to notify users that they are interacting with a machine, while deepfakes and any synthesized content must receive special labeling. This applies not only to industry giants but also to all players operating in the European market.
Key Changes and Deadlines
Obligations for GPAI providers formally took effect in August 2025, but only now does the regulator have the tools for their enforcement. Companies are required to document the technical parameters of their models, disclose data to downstream providers, and comply with EU copyright rules. For the most powerful systems with systemic risks, an additional requirement is stipulated—notifying the European AI Office and conducting risk assessments.
New providers entering the market after August 2, 2025, are already required to comply with the rules. For others, the deadline extends to August 2, 2027.
Fines: Two Regimes, Different Thresholds
Penalties are divided into two levels. For violations of GPAI provider obligations (Article 101), a fine of up to €15 million or 3% of global annual turnover is stipulated. This applies to refusal to provide documents, submission of false data, or failure to meet model evaluation requirements.
A stricter regime—under Article 99—applies to prohibited practices, such as social scoring or manipulative systems. Here, fines reach €35 million or 7% of turnover. For providing incomplete information to national regulators—up to €7.5 million or 1% of turnover. A lower threshold applies to small and medium-sized businesses, while a higher one applies to large corporations.
Transparency and Labeling
The new rules require:
- explicitly disclosing interaction with AI during direct communication;
- informing about emotion recognition and biometric categorization systems;
- disclosing the use of AI in any content;
- adding machine-readable labels to synthesized materials.
For generative systems released before August 2, 2026, a transitional period applies until December.
I would note separately: the voluntary GPAI Code of Practice, signed by Anthropic, Google, Microsoft, OpenAI, and others, does not replace the law. Interestingly, xAI has only agreed to the safety chapter—meaning it will have to demonstrate compliance with transparency and copyright rules through other means. This sets a precedent for different compliance strategies.
High-Risk Systems and Outlook
Rules for high-risk AI systems will only take effect on December 2, 2027, and for AI in physical products—on August 2, 2028. The delay is linked to the Digital Omnibus and the need to prepare standards. The regulator has already launched tools for filing complaints and for whistleblowers.
As an analyst, I see a clear signal here: the EU is moving from soft regulation to strict control. Companies operating in Europe should already review their documentation and labeling processes now to avoid fines that could reach hundreds of millions of dollars for major players.