An agentic framework is software that coordinates one or more AI agents so they can break a goal into steps, choose actions, use tools, and revise their approach as conditions change. Instead of a single prompt-response loop, it manages planning, memory, task routing, and tool execution across a workflow.
In cyber security, this matters because agentic systems can speed up both offense and defense. Attackers may use them to automate reconnaissance, credential testing, log sorting, or multi-step intrusion workflows after initial access. Defenders use similar frameworks for alert triage, enrichment, and incident response playbooks. The security risk comes from granting agents broad permissions: if they can access repositories, secrets stores, cloud APIs, or admin tools, a mistake or compromise can expand quickly. Strong controls include least privilege, approval gates for high-risk actions, audit logging, and tight separation between data sources and actuation.