Capability Narrowing
A security property in multi-agent collaboration stating that when an agent delegates a subtask, the permissions obtained by the recipient must be a subset of those currently held by the delegator rather than a union or superset. It requires verifiable identity, task scope, and a delegation chain record.
Timeline (last 90 days)
Capability Narrowing requires: when Agent A delegates a subtask to Agent B, the permission scope B receives must be a subset of the permissions A currently holds, not a union or a superset.
If a link in the delegation chain attempts to expand its permissions, the authorization system should directly reject the request, so that even if an intermediate Agent is compromised, the attacker cannot obtain permissions beyond the original task scope through the delegation chain.
All Facts (2)
Capability Narrowing requires: when Agent A delegates a subtask to Agent B, the permission scope B receives must be a subset of the permissions A currently holds, not a union or a superset.
50%UnverifiedIf a link in the delegation chain attempts to expand its permissions, the authorization system should directly reject the request, so that even if an intermediate Agent is compromised, the attacker cannot obtain permissions beyond the original task scope through the delegation chain.
50%