EU Begins Enforcing AI Act Transparency and Safety Rules

The European Union began enforcing key provisions of the AI Act on August 1, 2026. These measures introduce strict transparency requirements, mandating that chatbots identify themselves as AI and that AI-generated or altered media, including deepfakes, be labeled with machine-readable markings to prevent deception.

The European Commission AI Office now possesses the authority to regulate advanced general-purpose AI models. This includes the power to test models for compliance, demand access to systems before they enter the market, and restrict the deployment of AI that poses unaddressed cybersecurity or fundamental rights risks. Providers of these models must also document training content and implement copyright policies.

Certain AI practices are now prohibited, including social scoring, predictive policing, human behavior manipulation, and emotion recognition in workplaces or schools. Companies failing to comply face significant penalties, with maximum fines reaching 35 million euros or 7% of their annual global turnover.

Regulatory implementation will continue in phases. A ban on sexualized deepfakes is scheduled for December 2026, while requirements for specific high-risk AI systems are postponed until December 2027 and August 2028.