EU publishes final AI transparency Code ahead of Article 50

A voluntary playbook for labelling AI-generated content landed in Brussels this month, and the timing matters more than the announcement. The European Commission published its final Code of Practice on Transparency of AI-Generated Content in June 2026, weeks ahead of the AI Act's Article 50 transparency obligations becoming applicable on August 2, 2026, according to TechPolicy.Press. Organizations have until July 22 at 18:00 CET to submit a signatory form if they want to be on the initial list.

The split between provider and deployer obligations is the part to internalize. Providers of AI systems generating synthetic content must implement machine-readable marking, detection mechanisms for verification, and quality requirements covering effectiveness, reliability, robustness, and interoperability. Deployers, the people putting the systems to use, have to disclose deepfakes and AI-manipulated text via harmonized EU labels or equivalent mechanisms, and make those disclosures clear, accessible, and visible at first exposure. Exemptions exist for artistic, creative, satirical, and fictional works, plus human-reviewed editorial content.

The reach is extraterritorial. The explainer notes that providers and deployers fall within scope even when established outside the EU, provided the AI system output is used in the Union. Non-compliance can run up to EUR 15 million or 3% of worldwide annual turnover, which is what makes a voluntary code worth signing. Adherence buys you a known posture against an enforcement regime that has not yet been tested in practice.

The honest caveats are in the framework itself. It acknowledges that no single marking technique meets all four requirements simultaneously, that technical benchmarks remain underdeveloped, and that definitional gaps, including what counts as a deepfake versus ordinary editing, will be worked out case by case during enforcement. What the explainer does not give you is how regulators will weigh signatory status when penalty decisions actually start landing, or how online platforms hosting third-party content, which the Code says generally are not considered deployers, will be treated when the disclosure burden falls on someone else.

For teams shipping generative tools into the EU, the next month is the practical window: decide whether to sign by July 22, and have a marking and disclosure story in place before August 2.