Conceptuncovered capability map / 工具能力地图
Capability Map
A concept in AI Agent security that refers to a mapping framework classifying all tools callable by an agent according to their actual business capabilities rather than function names. This framework is used to identify blind spots in permission controls and potential bypass paths that adversaries or the agent itself might exploit to circumvent access restrictions.
Timeline (last 90 days)
Oct 9
Controlling only by function name or output wording leaves unclosed bypasses (capability maps not covered by the policy).
Unverified50%
Oct 9
Capabilities should be grouped according to what an action can change, not according to function names.
Unverified50%