The EU AI Act Is Now Being Enforced. Here Is What That Means

The EU AI Act stopped being theoretical on August 2. The European Commission's AI Office is now actively enforcing transparency rules for AI systems across all 27 member states. If you run a chatbot, image generator, or content tool that serves European users, you have legal obligations you did not have a month ago.

What is being enforced now

Three categories of transparency rules are live. Chatbots must clearly identify themselves as AI when interacting with humans. AI-generated deepfakes, whether video, audio, or image, must carry visible labels. And machine-generated content must include machine-readable watermarks so platforms and fact-checkers can detect it programmatically.

The AI Office handles enforcement for general-purpose AI models directly. National authorities in each member state cover everything else, from hiring algorithms to content moderation tools.

The penalty structure

Non-compliance carries fines of up to 15 million euros or 3% of global annual turnover, whichever is higher. For context, a company generating 10 billion euros in annual revenue could face a fine of 300 million euros. The structure mirrors GDPR's proportional approach: designed to hurt regardless of company size.

What comes next

The AI Act rolls out in stages. From December 2026, it will ban AI systems that generate non-consensual explicit content or child sexual abuse material. High-risk AI system rules, covering biometric identification, critical infrastructure, and employment decisions, were pushed back to December 2027.

That delay gives companies roughly 16 more months to prepare for the most complex compliance requirements. But the transparency rules active today are not optional warm-ups. They carry full enforcement power and full financial penalties.

We explored how AI model fingerprinting could help with compliance in our piece on model genome fingerprinting. The EU Act is now the regulatory framework those technical solutions were built for.