- AI system
- A machine-based system that infers from its input how to generate outputs such as predictions, content, recommendations or decisions.
- Provider
- Develops an AI system, or has it developed, and places it on the market or puts it into service under its own name.
- Deployer
- Uses an AI system under its own authority in a professional setting.
- General-purpose AI model
- A model that can perform a wide range of tasks and can be built into many downstream systems.
- High-risk AI system
- A system in a category the EU AI Act lists, or a safety component of a regulated product. The strictest requirements apply.
- Prohibited practice
- A use of AI that the EU AI Act bans outright.
- Conformity assessment
- The process that shows a high-risk system meets the requirements before it goes on the market.
- Fundamental rights impact assessment
- An assessment that certain deployers carry out before first using a high-risk system.
- Human oversight
- Measures that let people understand, monitor and, where needed, override or stop a system.
- Post-market monitoring
- The provider's system for collecting and reviewing how a system performs once it is in use.
- Serious incident
- A malfunction or failure with grave consequences, which must be reported.
- AI management system
- The policies, roles and processes an organisation uses to govern AI, as set out in ISO/IEC 42001.
- Model card
- A short document on a model's purpose, its performance and its limits.
- Bias
- Systematic error that produces unfair outcomes for a group of people.
- Explainability
- The ability to give the reasons behind an output in terms a person can follow.
- Data drift
- A change in the input data over time that lowers a model's performance.
- Red teaming
- Structured adversarial testing to find failures before others do.
- Hallucination
- An output of a generative model that is fluent and wrong.