An agent session is a bounded period of AI activity in which an agentic system carries out tasks, calls tools, reads files, or exchanges messages on behalf of a user or workflow. Unlike a single chat turn, a session can include multiple steps and side effects, making it useful for tracking what the system saw and did.
In cyber security, agent sessions matter because they create an auditable trail. Security teams can review session metadata, prompts, tool invocations, and file access to detect abuse, verify policy compliance, and support incident response. In attacks, a compromised or malicious session may be used to automate reconnaissance, exfiltrate data, or trigger unsafe actions across connected systems. In defenses, session logging helps correlate AI activity with identities, permissions, and downstream events, giving organizations a clearer view of how autonomous or semi-autonomous AI was used.