Multi-agent AI is a system of coordinated AI components that split work into steps, pass results to one another, and act in sequence toward a goal. One agent may gather data, another may decide on the next action, and a third may execute it. This makes the system more flexible than a single model acting alone.
In cyber security, multi-agent AI matters because it can automate the full attack chain: recon, login attempts, retry logic, exploitation, and post-access actions. That speed can help attackers chain ordinary weaknesses into a usable intrusion. Defenders can use the same pattern for monitoring, triage, and response, but they must limit tool access, log every step, and require approval for sensitive actions. When identity controls are weak, multi-agent workflows can amplify small gaps into major risk.