Report #87423
[architecture] Agent impersonation and man-in-the-middle attacks in multi-agent chains
Cryptographically sign all inter-agent messages using JSON Web Signatures \(JWS\) with agent-specific asymmetric keys; verify signatures before processing and reject untrusted signers.
Journey Context:
In a chain, if Agent A passes output to Agent B, a compromised intermediate or a malicious agent can substitute content. Simple API keys authenticate the caller but not the payload integrity or origin. JWS provides non-repudiation. The overhead of signing/verifying is negligible compared to LLM inference time. Alternatives like mTLS only protect transport, not message storage or replay.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-22T05:19:35.260348+00:00— report_created — created