The EU's AI Act Now Requires Labeling AI-Generated Content Starting August 2
Generative AI providers must watermark their output. Deployers must disclose deepfakes and label AI-generated text on matters of public interest. Fines run up to 15 million euros.
Starting August 2, AI-generated content in the EU needs a label. Here is what actually counts.
Starting August 2, 2026, Article 50 of the EU AI Act requires generative AI providers to mark their output as AI-generated and requires anyone deploying that content to disclose it clearly, with fines running up to 15 million euros or 3 percent of global turnover.
Who this actually binds
Two separate obligations. Providers, the companies building the generative AI system, must make their text, audio, image, and video output carry a technically detectable, machine-readable mark. Deployers, anyone publishing or using that output, must clearly disclose deepfakes and must label AI-generated or manipulated text on matters of public interest, unless a human reviewed it and takes editorial responsibility for it. The rule reaches outside the EU too: any provider or deployer whose product or output targets EU users is bound, regardless of where the company is based.
What counts as public interest text
The text labeling duty has a real carve out. Content that got substantive human editorial review, with an identifiable person or organization taking responsibility for it, does not need a label even if AI helped write it. Obvious synthetic content, assistive editing tools, law enforcement use, and clearly marked artistic, satirical, or fictional work get lighter touch or no disclosure duty. A voluntary EU Code of Practice on Transparency of AI-Generated Content offers a path to a compliance presumption, but the underlying legal duty applies whether or not a company signs onto it.
The transition window
Systems already on the market when the rule takes effect get until December 2, 2026 for the machine-readable marking duty specifically. Everything else starts August 2.
Why a build studio cares
Any AI workflow we ship that touches EU users now has a labeling requirement baked into the output, not an afterthought. Worth checking this at the design stage of an AI feature, not the compliance review stage, since retrofitting a watermark or a disclosure flow after launch is more work than building it in from the start.
Next step: read the European Commission's overview or the Article 50 explainer. If an AI feature you're building touches EU users, write to us at hello@gattyworks.com.