Agentic techniques are methods in which an AI system or other software uses tools, follows workflow steps, or delegates actions to complete a task, rather than only generating text. Examples include querying an internal database, reading documents, or triggering an operation through a connected service.
In cybersecurity, these techniques matter because connected tools can expose sensitive data or perform consequential actions. Attackers may misuse an agent’s access, workflows, or delegated permissions; defenses include limiting tool access to what is necessary, requiring approval for sensitive actions, and logging activity. The term alone does not establish that an autonomous AI agent was compromised or explain how an intrusion occurred.