Disclosure · User awareness
Transparency obligations
Sometimes the main risk of an AI system is simply that you don't know it's there. For those situations the AI Act prescribes neither bans nor heavy engineering — it prescribes honesty.
01What transparency means for users
In defined situations, people must be told that AI is in play: when a system converses so naturally it could pass for a person, and when images, audio or video are synthetic. Disclosure restores the user's footing — you evaluate advice differently when you know its source.
02The classic situations
- Chatbots — a public office's always-available assistant must not masquerade as a human clerk. The game's case "The desk that always answers" turns exactly on this plus the reliance it creates.
- Synthetic content and deepfakes — generated images and video must be identifiable as such; "The synthetic city" case explores a municipality flooding its channels with generated scenes.
03Why this is a citizenship issue
Transparency is the cheapest safeguard and the most democratic one: it lets ordinary people apply their own judgment. It connects directly to AI literacy (can you spot the machine?) and digital citizenship (do you have the right to know?).
04The limits of a label
The productive classroom question is when disclosure is not enough. A label on a chatbot does not fix wrong answers relied on by vulnerable users; a watermark does not undo a viral deepfake. Transparency is a floor, not a ceiling — which is why some systems climb into the high-risk tier no matter how honest their labels are. The transparency lesson plan stages this debate.
05Who has to disclose what
Transparency is not one undifferentiated duty: it changes with what the system does and who puts it into use. This is the simplified reading we use in class — the reference text is Article 50 of Regulation (EU) 2024/1689.
| Situation | Who carries the duty | What the person must be able to know |
|---|---|---|
| System that interacts directly with people | Whoever provides the system | That they are interacting with an AI, unless it is already obvious to anyone paying ordinary attention. |
| System generating synthetic image, audio, video or text | Whoever provides the system | That the content is artificially generated: the marking must be machine-readable, not just visible to the human eye. |
| Deepfakes published to the public | Whoever publishes or spreads them (deployer) | That the content was artificially generated or manipulated. |
| Emotion recognition or biometric categorisation | Whoever uses the system (deployer) | That the system is running on them — assuming that use is allowed at all: many applications in schools and workplaces are prohibited. |
Two consequences worth pointing out in class. First: the duty can move along the chain — whoever builds the generator marks the output, whoever publishes it discloses. The guide to providers and deployers maps that distinction and explains when a role can change. Second: "obvious" is not a convenient escape hatch, because it is measured against a reasonably attentive person, not an expert user.
On timing: transparency obligations apply from the regulation's general application date, 2 August 2026. The full schedule is on the application timeline.
06Exercise: is the label enough?
Three situations to classify in pairs, five minutes. There is no clean answer: the discussion is worth more than the box.
| Scenario | One possible reading |
|---|---|
| A city council publishes a fully generated promotional video with a small watermark in one corner. | The disclosure exists, but the real question is whether it is noticeable: a marking nobody sees satisfies the form and misses the purpose. |
| A health chatbot introduces itself as an AI and then advises ignoring a symptom. | Transparency is respected here but is not enough: use in a healthcare setting pushes the system towards far heavier obligations than mere disclosure. |
| A council assistant never says it is an AI, but sits on a page titled "Automated assistant". | Context can make the AI obvious. The question to ask in class: obvious to whom? To a computer science student, or to an elderly person at the counter? |
The case "The desk that always answers" in the game puts the player in exactly this position: the disclosure is present, people's reliance is misplaced anyway, and the inspection report has to hold both facts together.