EU Begins Enforcing AI Act With New Powers Over Advanced Models

The European Union has started enforcing its AI Act, introducing new regulatory measures for advanced artificial intelligence systems and establishing penalties for non-compliance.

From August 2026, EU authorities have the ability to request information, impose fines, and order restrictions or recalls on AI models if risks are identified. The enforcement follows the AI Act’s entry into force in August 2024, with general application and oversight measures becoming effective two years later.

Transparency requirements under Article 50 now require providers to label AI-generated content and inform users when they are interacting with AI systems. These obligations apply to general-purpose AI models, including those used in large online platforms.

EU cybersecurity agencies have been granted access to advanced AI models such as OpenAI’s GPT-6-ASTRA and Anthropic’s Mythos 5 for the purposes of testing and risk assessment. This access is part of the broader effort to monitor and manage potential risks associated with autonomous AI agents.

What the numbers show

• The AI Act allows fines up to €15 million or 3% of global turnover for general-purpose AI violations.
• For prohibited AI practices, fines can reach €35 million or 7% of global turnover.
• The Act entered into force on 1 August 2024, with enforcement starting on 2 August 2026.

The AI Office and national authorities share responsibilities for enforcement. The AI Office oversees general-purpose AI models and those integrated into very large online platforms, while national authorities are responsible for other AI systems.

On 11 September 2026, the European Union called on technology companies to ensure their advanced AI models are properly managed. This statement followed a series of security incidents involving autonomous AI agents.

The AI Act’s enforcement powers include the ability to demand detailed information from providers, apply sanctions, and require the withdrawal of AI models from the market if necessary. These measures are intended to address risks as they arise and ensure compliance with new transparency and safety standards.

Administrative penalties under the AI Act vary depending on the type of infringement. Providers of general-purpose AI models face fines of up to €15 million or 3% of annual worldwide turnover, while breaches involving prohibited AI practices can result in higher penalties.

* This article is based on publicly available information at the time of writing.