EU publishes Article 50 transparency rules days before enforcement
Providers must tell users they are talking to an AI and mark synthetic content as machine-readable. Deployers must disclose deep fakes and biometric systems.
- The European Commission published guidelines on Article 50 transparency obligations, which begin applying August 2, 2026.
- Providers must inform users when they are interacting with an AI system and embed machine-readable marks on synthetic content.
- Deployers must disclose deep fakes, AI-generated public-interest publications, and emotion-recognition or biometric-categorisation use.
On July 20, 2026 the European Commission published guidelines on the transparency obligations for providers and deployers of certain AI systems under Article 50 of the AI Act, whose rules begin applying August 2, 2026. The timing is the story. The guidance landed less than two weeks before the obligations take effect, giving vendors a narrow window to read the Commission's interpretation and adjust.
What providers and deployers must do
The guidelines split duties between the two roles. Providers must inform users when they are interacting with an AI system, and they must embed machine-readable marks so that synthetic content can be detected as AI-generated or manipulated, the Commission said. Deployers carry a separate set of disclosures. They must reveal deep fakes, flag AI-generated publications on matters of public interest, and disclose the use of emotion-recognition or biometric-categorisation systems.
The Commission said the guidance aims for consistent, proportionate, and uniform application across the bloc, and it arrives alongside a related Code of Practice on transparency, per the Commission's digital strategy office. The guidelines are not themselves binding law, but they signal how regulators intend to read the statute, which is what compliance teams actually plan against.
The distinction between marking and labelling runs through the text. Marking is the machine-readable signal embedded in the content itself, the part a detection tool reads. Labelling is the human-facing notice that a piece of media is synthetic. An agent that generates text, images, or audio for European users is expected to do both, and the Commission's framing makes clear it views the two as complementary rather than interchangeable.
The compliance clock, in two dates
There are effectively two deadlines to track. The core obligations begin applying August 2, 2026. Separately, AI systems already placed on the market before August must comply with the labelling and tracking requirements from December 2, 2026. That staggering gives existing products a few extra months to retrofit content marking, while anything shipping fresh in August is expected to comply from the start.
What this means for the agent economy
Interactive-AI disclosure and machine-readable content marking land squarely on the products GaaS vendors sell. A conversational agent is, by definition, an interactive AI system a user might mistake for a human, so the "tell users they are talking to an AI" rule is not an edge case for this industry, it is the industry. Content-generating agents inherit the marking obligation on everything they produce. For any vendor serving European users, these are concrete engineering tasks with a date on them, not abstract principles.
This is the operational detail underneath the broader enforcement wave we tracked in our coverage of the August 2 AI Act obligations for agents. The guidance narrows the ambiguity about who must do what, which is welcome, but it also removes the excuse of uncertainty. Vendors that had been waiting to see how Article 50 would be read now have their answer, and roughly ten days to act on it.