AI-agent control means an autonomous AI system directed or selected actions during a process, rather than merely being present or used as a tool. In cyber security, this is a specific claim about how activity was controlled; rapid actions alone do not establish it. A human operator or conventional script can also respond quickly.
In an attack investigation, evidence may include reliable records of agent decisions, tool calls, and the actions those calls triggered, correlated with identity, application, endpoint, and network logs. Defenders should preserve these records and distinguish confirmed behavior from hypotheses. A quick change in tools after a failed attempt may warrant investigation, but without supporting telemetry it does not prove an AI agent was in control.