EU AI Act Article 50 transparency rules enter force

Details

Article 50 of the EU AI Act entered force in August 2026, imposing transparency and disclosure obligations on both AI providers and deployers across the bloc. Providers must design systems to notify users when interacting with AI rather than a human, with narrow exceptions for obvious cases and law enforcement contexts that include third-party rights safeguards. The content-generation obligation requires machine-readable marks on AI-generated synthetic audio, image, video, and text — with a proportionality test balancing implementation cost against the technical state of the art. Assistive editing (a minor photo touch-up) falls outside the requirement, but wholesale AI-generated content replacements do not. Deepfakes face their own disclosure duty for deployers. This coverage adds detail beyond standard reporting: biometric categorization and emotion recognition systems trigger separate disclosure requirements, and personal data collected through those systems still falls under GDPR, the EU institutions data protection regulation, or the Law Enforcement Directive depending on context. The Act's marking requirement must be effective and interoperable, not just technically present — cheapest implementation alone will not satisfy it.