An autonomy benchmark is a test that measures how independently an AI system can complete multi-step work without human guidance. In cyber security, these benchmarks are used to see whether a model can keep track of a task, recover from mistakes, and finish a sequence of actions that resembles real operational work.
This matters because many attacks are not single-step exploits. They involve repeated decisions such as reconnaissance, privilege escalation, lateral movement, and objective completion. A model that scores well on an autonomy benchmark may be able to support longer, more complex offensive workflows, but the same capability can also help defenders automate triage, detection, and response. The key security question is not whether an AI can answer a prompt, but whether it can sustain useful action across a chain of steps. For that reason, autonomy benchmarks help teams judge both attack risk and the value of response automation.