On Aug. 2 the European Commission’s AI Office begins actively enforcing the EU AI Act, activating Article 50’s transparency regime and the power to investigate providers, order corrective measures, and levy fines of up to €15 million or 3% of worldwide annual turnover. It’s the first slice of the Act with real teeth, and it lands on a market that hasn’t prepared for it.

The Article 50 obligations are narrow but concrete. Chatbots and interactive systems have to tell users they’re talking to AI, not a human. Deepfakes need labels. Generative outputs need machine-readable watermarking. Systems already on the EU market before Sunday get until Dec. 2 to implement watermarks, and deepfakes produced before Aug. 2 are exempt from retroactive labeling per Tech Policy Press. The Commission has also published three voluntary AI-content icons. Chatbot and deepfake disclosure rules carry no grace period.

The compliance posture is worse than Brussels likely hoped. A Vision Compliance analysis cited by TechTimes found roughly 78% of organizations had taken no meaningful AI Act compliance steps as of spring 2026. The GPAI Code of Practice, the voluntary on-ramp, was signed by 24 firms by June, including Amazon, Anthropic, Google, IBM, and Microsoft. Meta declined outright. xAI signed only the Safety and Security chapter.

Enforcement is shared between the AI Office, national market surveillance authorities, and the European Data Protection Supervisor. Heavier obligations remain scheduled: Annex III high-risk systems on Dec. 2, 2027, and Annex I embedded systems on Aug. 2, 2028.

Brussels has spent years being told its AI rulebook was unenforceable vibes. Sunday it starts writing the fines.

Sources

Sources