The European Union began enforcing the AI Act's Article 50 transparency obligations on August 2, 2026, requiring companies to disclose when users interact with AI and when content has been generated or altered by AI systems. The rules apply immediately to new AI systems, while models and services launched before that date have a four-month grace period until December 2, 2026 to achieve compliance. The Verge

The obligations draw a structural line between two roles. Providers — companies that develop and market AI systems — must design those systems to explicitly notify users when they are interacting with AI rather than a human, unless the AI nature is already obvious from context. Providers must also embed machine-readable marks on synthetic audio, image, video, and text output. Deployers, defined as platforms and services that use AI systems, carry a complementary duty: they must label any AI-generated or manipulated image, audio, and video deepfake content that is designed to look real, and inform people when they are exposed to AI-generated content. Some companies will be classified as both provider and deployer. The Verge · European Commission AI Act Service Desk

Non-compliance carries financial teeth. Companies that fail to meet the transparency requirements risk fines of up to 15 million euros (approximately $17.2 million) or 3 percent of their global annual turnover, whichever is higher. The Verge

To support implementation, the European Commission has published a package of guidance materials. Guidelines on the transparency obligations for providers and deployers of certain AI systems were released on July 20, 2026. European Commission A Code of Practice on Transparency of AI-generated Content followed, with its obligations applying from August 2. European Commission The Commission also developed a set of optional AI disclosure labels — standardized icons that platforms can adopt to signal AI-generated content in accordance with the Act. The labeling requirements themselves are mandatory; only the specific Commission-designed labels are optional. European Commission

The Commission's rationale is straightforward: the rapid development of generative and interactive AI systems has made it increasingly difficult for people to distinguish AI interactions and AI-generated content from human-created and authentic material. Chatbots and other interactive AI systems must now tell users they are dealing with AI rather than a human. European Commission

Article 50 sits within the broader risk-tiered architecture of the AI Act, which defines four levels: unacceptable risk, high risk, limited/transparency risk, and minimal or no risk. The transparency obligations fall under the "limited risk" tier — systems permitted on the market but subject to disclosure duties. The Act imposes no rules on AI deemed minimal or no risk. European Commission

The enforcement timeline follows a staggered schedule. Prohibitions on eight categories of AI practices — including harmful AI-based manipulation and deception, social scoring, untargeted scraping of internet or CCTV material to build facial recognition databases, emotion recognition in workplaces and education institutions, and real-time remote biometric identification for law enforcement in public spaces — took effect in February 2025. A ninth prohibition, covering AI systems that generate non-consensual sexually explicit and intimate content or child sexual abuse material, comes into effect in December 2026. It was introduced as part of the AI Omnibus. Starting December 2, 2027, high-risk AI systems will face strict pre-market obligations including risk assessment and mitigation systems, high-quality datasets, activity logging for traceability, detailed documentation, clear information to deployers, human oversight measures, and robustness, cybersecurity, and accuracy requirements. European Commission

The broader context here is that the transparency tier is the first part of the AI Act to touch the generative AI stack directly. The prohibited-practices list, enforced since early 2025, primarily addressed surveillance and social-control applications. Article 50 reaches into the output layer of every chatbot, image generator, and synthetic media tool serving EU users, and it does so through two distinct enforcement surfaces: a provider-side machine-readable marking requirement embedded in the generation pipeline, and a deployer-side labeling requirement applied at the point of content presentation. For a company operating both a foundation model and a consumer-facing interface, that means implementing watermarking or provenance signals at the model level and visible disclosure at the product level.

Worth flagging is the interaction between the four-month grace period and the technical work involved. Embedding machine-readable marks in synthetic output is not a configuration toggle for most production systems; it requires changes to the generation pipeline, potentially to model architecture or post-processing layers. Companies that deferred implementation planning until the enforcement date now have until December 2 to retrofit systems already in production — a window that may prove tight for providers with large deployed model fleets.

The optional Commission-designed labels address a different concern: fragmentation. Without standardized iconography, each platform would develop its own visual language for AI disclosure, producing the kind of inconsistent UX that undermines the purpose of a transparency mandate. Whether major platforms adopt the Commission's labels or develop their own variants will determine whether users encounter a consistent disclosure experience across services or a patchwork of platform-specific badges. That adoption picture should become clearer over the coming months.