What are the Obligations Under the EU AI Act?

The obligations for AI systems under the AI Act are determined based on their risk categorization. “High-risk” AI systems are subject to rigorous assessments for accuracy, robustness, and cybersecurity before they can be deployed, while the providers of such systems must maintain detailed documentation, conduct lifecycle monitoring, and ensure absolute transparency for users whenever they interact with the system.

Any AI system that has the potential to carry significant implications for the health, safety, or fundamental rights of its users can be defined as “high-risk”. The most prominent examples are AI systems used in critical infrastructure, employment and workers management, essential private and public services, law enforcement, migration, asylum, and border control management. Any AI system determined to be “high-risk” is subject to strict compliance obligations owing to their potential impact on individuals as well as society in general.

AI systems that have been determined to carry minimal risk are subject to similarly minimal obligations. These include basic transparency measures such as ensuring users are aware when they’re interacting with AI and correct labeling of content produced by AI.