Brussels gains powers to test and block advanced AI models
The EU’s AI Office will now have the authority to audit and suspend advanced artificial intelligence models, exposing tech companies to strict new compliance burdens and massive fines.
From Sunday, the European Union will begin enforcing its AI Act, granting a new regulatory body the power to audit, restrict, and potentially block advanced artificial intelligence models. The AI Office, staffed with dozens of tech experts, lawyers, and economists, can now force companies to hand over their most powerful general-purpose systems for testing before or after they enter the European market. If a developer fails to address regulatory concerns, Brussels can restrict the model's deployment across the bloc.
This shift from legislation to active policing fundamentally changes the risk profile for tech companies operating in Europe. Earlier this year, the EU struggled to gain access to Anthropic's Mythos model, but regulators will no longer have to rely on voluntary cooperation. Businesses must now navigate a mandatory access regime alongside strict transparency rules requiring them to label AI-generated deepfakes and disclose when users are interacting with a chatbot.
For investors and executives, the financial stakes of non-compliance are severe. Companies that deploy banned practices face fines of up to 7% of their global annual turnover or €35 million, whichever is higher. These banned practices include predictive policing, emotion recognition in schools or workplaces, and systems that manipulate human behaviour.
Other breaches, including obstructing EU probes, carry penalties of up to 3% of global revenue or €15 million. Existing AI systems have until December 2 to adapt to the framework, though exemptions exist for artistic, satirical, or fictional works. While national authorities will oversee smaller systems, the AI Office retains direct control over the most advanced models.
Additional restrictions are scheduled to take effect later this year, including a December ban on AI-generated sexualised deepfakes following outrage over non-consensual explicit images produced by Elon Musk's chatbot Grok. However, Brussels has delayed its strictest high-risk requirements to give companies more time to prepare. Rules for stand-alone high-risk AI systems will apply from December 2027, while AI tools embedded in other products have until August 2028.
