Autonomous agents are software systems that can plan and execute a sequence of tasks with limited human intervention. They usually combine a model with tools, memory, and rules so they can move through a workflow: gather data, make decisions, call APIs, and adjust based on results. In cyber security, that makes them useful, but also risky, because a mistake can propagate quickly across many steps.
Defenders use autonomous agents for alert triage, log analysis, vulnerability prioritization, and routine response actions in SOAR-style workflows. Attackers can use the same pattern to scale reconnaissance, phishing, credential testing, or malware operations. Security teams care about agent controls such as least privilege, approval gates, input validation, and monitoring, since prompt injection, poisoned data, or overly broad tool access can cause an agent to leak data or take unsafe actions.