Risk-based approach: four tiers, one logic

The AI Act's core structure is a four-tier risk ladder: unacceptable risk is banned outright (e.g., social scoring by governments), high risk faces strict controls like conformity assessments and human oversight, transparency obligations apply to limited-risk systems such as chatbots that must disclose they are AI, and minimal risk is free from regulation. The key contrast is that high-risk systems require pre-market checks, while transparency systems only need post-market disclosure—a difference exam questions often blur by swapping the duties.

To avoid traps, remember that high-risk always involves a mandatory conformity assessment before deployment, whereas transparency obligations are lighter and apply after release. For example, an AI used for recruiting is high-risk and needs a notified body check; a chatbot only needs to tell users it is AI. Another trick: if a question mentions "limited risk" or "transparency," look for duties like labelling or informing, not full pre-market approval. Also, note that general-purpose AI (GPAI) has its own rules, separate from the four tiers, so don't mix them.

A compact way to test yourself: name one banned use (e.g., real-time biometric surveillance in public), one high-risk use (e.g., credit scoring), and one transparency use (e.g., deepfake disclosure). If you can list three distinct examples without mixing duties, you have the logic locked. The exam will try to make you apply the wrong tier to a scenario—stay anchored to the pre-market vs. post-market split.