The EU AI Act's next practical deadline is not just for companies building the riskiest AI systems. Article 50 transparency rules start applying on Aug. 2, 2026, and they can affect chatbots, AI agents, synthetic media tools, deepfake workflows and biometric or emotion-recognition systems used in professional settings.
The short version: if a person is directly interacting with AI, looking at AI-generated or manipulated media, or being exposed to certain biometric tools, someone may need to tell them clearly. The European Commission published fresh guidance on July 20 to explain who has that job and when the notice has to appear.
This is a compliance starting point, not legal advice. If your product, marketing workflow, newsroom, agency, HR team or platform touches EU users, use the next few days to check these four disclosure questions.
1. Is a person talking directly to AI?
Providers of AI systems that directly interact with people must design them so users are informed they are interacting with AI, unless that fact is obvious. The Commission's Q&A points to chatbots, AI agents and avatars as examples, and says the notice should appear from the start of the first interaction in a clear and accessible way.
2. Does your system create synthetic content?
Providers of systems that generate synthetic audio, image, video or text must add machine-readable marks that make the output detectable as AI-generated or manipulated. The guidance describes exceptions, including standard editing and some machine-to-machine or closed industrial uses, but the safe first step is to map where generated outputs leave your internal workflow and reach people.
3. Are you deploying deepfakes or public-interest AI text?
Deployers have a separate duty. Deepfake image, audio or video content generally needs a clear disclosure when a person first sees or hears it. AI-generated or manipulated text published to inform the public on matters of public interest also needs labeling unless it went through human review or editorial control and a person or organization takes editorial responsibility.
4. Are biometric or emotion tools involved?
Deployers of emotion-recognition or biometric-categorization systems must inform people exposed to those systems. That can matter for workplace, retail, education, event, security and advertising uses, even when the organization did not build the underlying model.
What to do this week
Make a short inventory before Aug. 2: every AI-facing customer touchpoint, every synthetic-content output, every AI-edited public communication, and every biometric or emotion-analysis use. For each item, write down whether your organization is the provider, the deployer or both; what notice users see; when they see it; and who approved the wording.
One grace period is narrow. The Commission says Article 50 applies from Aug. 2, while some systems already on the market before that date get until Dec. 2, 2026, only for the machine-readable marking and detection obligation for AI-generated content. That does not erase the broader disclosure work.
The penalty ceiling is serious: the Commission's quick-facts page says fines can reach 15 million euros or 3% of worldwide annual turnover for companies, with proportionality considered for smaller businesses. The better first move is not panic. It is a written disclosure map, a named owner and a decision on which AI outputs need a clear label before they reach people.
