The European Commission’s AI Office has begun enforcing the AI Act’s rules for the most powerful AI models, Executive Vice-President Henna Virkkunen said at the International AI Summit in Dublin on October 14, 2026. The office has sent requests for information to AI developers covering security practices, outside testing and how their models are used.
What the AI Act requires
The AI Act, the European Union’s comprehensive framework for regulating artificial intelligence, treats the most powerful models as systemic-risk systems. Developers of these models must conduct and document external testing for dangerous capabilities, report serious incidents to the AI Office, and ensure their models cannot be used to produce unlawful content at scale.
Enforcement for the most powerful models was always going to be the hardest part of the AI Act. The rules were written to apply to a small number of frontier models, but the specifics of what counts as adequate testing, adequate incident reporting and adequate security have been debated since the law was passed.
What the AI Office is asking
Virkkunen said the office has sent requests for information on security, outside testing and model use. The first requests are information-gathering rather than enforcement actions, which is the standard regulatory sequence: establish what is happening, then determine whether it complies with the law.
The European Commission also published Implementing Decision (EU) 2026/2274, which lists EN 18286:2026 as the first harmonised standard supporting the AI Act’s rules for high-risk AI systems. The standard covers quality management systems for regulatory compliance, giving companies a concrete framework to follow.
How this affects US-based AI labs
The AI Act applies to any AI system placed on the EU market, regardless of where the developer is based. That means OpenAI, Anthropic, Google and Meta must all comply if their models are available in Europe. The requests for information sent in October 2026 are the first visible sign that the EU intends to enforce the rules against non-European companies.
The Trump administration has taken a different approach, naming a White House AI czar and pushing for voluntary industry agreements rather than binding regulation. That divergence creates a compliance burden for AI companies that must navigate both frameworks simultaneously.
For related coverage, see our report on OpenAI deprecating the GPT-4o family and Meta’s Muse Spark Open model.
Sources
- Oh My AI news archive, EU AI Office enforcement summary (October 14, 2026): https://ohmyai.ai/archive
- DW, “EU AI Office starts enforcing rules for most powerful models” (October 2026): https://www.dw.com/en/artificial-intelligence/t-42133476
