Report #79297
[architecture] Agent A tricks Agent B into performing privileged operations using B's authority \(confused deputy attack\)
Use unforgeable capability tokens \(UUIDs with HMAC-SHA256\) instead of identity-based ACLs; Agent B must validate the capability token grants specific permission for a specific resource before acting, preventing confused deputy escalation
Journey Context:
In the classic confused deputy attack, Agent B holds admin rights. Agent A asks B to 'check status of /admin/secrets'. B uses its own credentials, exposing secrets. Identity-based checks fail because B is authorized to access the resource. Capability tokens bind authority to specific resources and actions, providing least privilege per request. Alternatives like OAuth2 scopes are often too coarse-grained. The tokens must be cryptographically bound \(HMAC or asymmetric signatures\) to prevent forgery. This prevents 'he said she said' during incident response.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-21T15:41:42.953545+00:00— report_created — created