AI Glossary

The ability of a system to achieve correct and expected results in accordance with its intended purpose.

A Czech regulation referenced by the draft Czech AI Act in the exercise of market surveillance; for supervisory purposes, an AI system is considered a product.

A national piece of legislation that supplements a directly applicable EU regulation with the rules the regulation leaves to the member states, such as designating supervisory authorities, procedural rules and penalties.

A breach of the obligations laid down by the AI Act or national legislation.

Manipulating the inputs to an AI system in order to induce an erroneous or undesirable output.

A body composed of stakeholder representatives that provides the AI Office and the AI Board with expert and practical input.

A systematic review of an AI agent's activity, decisions and compliance with the rules, usually based on records of its actions.

Verifying the identity of an AI agent before granting it access to systems, data or tools.

A software structure providing the tools and building blocks for developing and deploying AI agents.

A set of rules, policies and controls governing the deployment, behavior and accountability of AI agents.

A repeated cycle in which an AI agent plans, calls tools, evaluates the results and continues until it has met the given goal.

The mechanism by which an agent retains information from previous interactions and uses it in subsequent steps or tasks.

Ongoing observation of an AI agent's activity, outputs and behavior for the purposes of oversight and the early detection of problems.

The coordination of multiple cooperating agents or tools so that they jointly solve a complex task.

A record of deployed AI agents containing their identity, permissions and defined tasks.

Measures protecting AI agents against misuse, unauthorized access and attacks, as well as preventing harmful conduct by the agent itself.

An advanced approach to AI in which the system autonomously decides how to proceed, uses tools and evaluates the results.

A workflow in which autonomous AI agents independently plan and carry out a sequence of steps to achieve a given goal with minimal human intervention.

Predefined conditions that an AI system must meet in order to be approved for deployment.

The principle of clearly assigning responsibility for the decisions, outputs and impacts of AI systems.

The set of measures ensuring that the use of AI complies with the requirements of the AI Act.

AI-based software capable of independently planning steps and carrying out tasks aimed at achieving a set goal.

Running an agent in an isolated environment with limited permissions, so that any errors or misuse do not cause harm beyond the defined space.

The intersection of the use of AI and consumer protection, covering for example the transparency of automated recommendations, the prohibition of misleading practices and liability for damage caused to consumers.

A programming interface providing access to artificial intelligence functionality from other applications.

Software using AI to support the completion of work or personal tasks through natural communication with the user.

An independent review of an AI system in terms of regulatory compliance, data quality, risks and impacts on the persons concerned.

The way an AI system is internally organized - how it connects perception, memory, reasoning, and action - in order to solve more complex tasks in a coordinated manner.

Tools and methods that attempt to identify whether a text, image or video was created by artificial intelligence; their reliability is limited and variable.

The obligation to clearly label artificially created or altered content so that users know it is AI output.