The European Union began enforcing Article 50 of the AI Act on August 2, 2026, requiring providers and deployers of certain AI systems across all 27 member states to label AI-generated content and disclose when users are interacting with an AI chatbot or agent rather than a human.

The European Commission announced the enforcement start in a press release dated July 31, 2026. The Article 50 rules coincide with the broader activation of most AI Act provisions on the same date, including those governing high-risk AI systems and general-purpose AI models (large models like GPT or Claude that can be adapted to many different tasks).

The Commission published detailed guidelines on transparency of AI-generated content on July 29, 2026, five days before enforcement began. It also maintains a Code of Practice on Transparency of AI-generated Content to help providers and deployers comply.

Under the new rules, two tiers of labeling apply. An "AI" mark is required when machines assisted in creating "authentic-looking" deepfake content — whether image, audio, video, or published text. Fully AI-generated content must carry an "AI-Generated" label. The Commission's examples include deepfake videos depicting politicians or fictional events, AI-composed music or art, and AI-generated news summaries. AI models that recognize emotions or process biometric data are also subject to labeling requirements.

The rules carve out exemptions. Personal content such as group chats is not covered. "Evidently artistic" satirical and fictional works are also exempt, a provision that should reduce friction for creative industries. The AI Act does not impose rules on AI systems deemed to pose minimal or no risk.

On the general-purpose AI model side, the European Commission will oversee providers, requiring them to document information, publish training data summaries, and implement copyright policies. National authorities handle enforcement of non-compliance with the transparency rules.

The financial exposure is meaningful. Companies that violate the transparency rules face fines up to €15 million or three percent of global annual revenue, whichever is higher. EU institutions, bodies, and agencies face a lower cap of €750,000.

Not all parts of the AI Act are moving on the original timeline. EU governments and lawmakers reached a provisional deal in May 2026 to delay rules on high-risk AI systems, including those involving biometrics and critical infrastructure. A retail association separately argued in June 2026 that AI-generated advertising should be exempt from the transparency rules. Neither effort has altered the Article 50 obligations that now apply.

For technology companies operating in the EU, the compliance surface is concrete and immediate. Any platform that surfaces AI-generated text, images, audio, or video to users must implement machine-readable and human-readable labeling. Chatbot and AI-agent deployments require disclosure that the user is interacting with an automated system, not a person. Providers of general-purpose AI models must maintain documentation, publish training data summaries, and have copyright policies in place.

The transparency rules complement, rather than replace, the AI Act's broader risk-based framework. High-risk system rules, general-purpose AI model rules, and now Article 50 transparency rules form three interlocking compliance tracks. Companies that have already built governance infrastructure for the high-risk provisions will find the transparency layer additive; those that have not now face the full stack simultaneously.

One open question is enforcement depth. The rules designate national authorities as the enforcement mechanism for non-compliance, but each member state's capacity and willingness to act will vary. The Commission oversees general-purpose AI model providers directly, which centralizes enforcement for the largest systems. For content-level transparency obligations, the practical test will be whether national regulators prioritize systematic non-compliance over individual violations.

The broader context here is that the EU is now the first major jurisdiction with enforceable, binding AI transparency requirements at this scale. The labeling framework creates a de facto content provenance standard — a way to trace where content originated and whether a machine was involved — that any global platform serving EU users must meet. Whether other jurisdictions align with, diverge from, or fragment around the EU model will shape the compliance architecture for AI-generated content worldwide. Companies building AI systems that produce or surface content should expect that provenance and labeling will become a baseline expectation, not a regional differentiator.