Report #39749
[architecture] Malicious or compromised agent impersonates another agent in the chain, injecting false data that downstream agents trust
Sign all inter-agent messages with the sender's private key using JWS \(JSON Web Signature\); downstream agents verify signatures against a registry of trusted public keys before processing; rotate keys regularly and revoke compromised agents immediately
Journey Context:
In closed systems, developers trust the network or use API keys in headers, but this fails when agents run on different infrastructure, when one agent is compromised, or when messages are replayed from logs. The alternative is mTLS \(mutual TLS\), which secures the transport but not the message content \(messages can be replayed from logs or inspected by middleware\). The right call is detached JWS \(RFC 7515\) on the payload, including timestamps \(iat, exp\) and unique IDs \(jti\) to prevent replay attacks. Tradeoff: Adds serialization overhead \(base64url encoding increases size ~33%\) and complex key management \(rotation, revocation lists\), but provides non-repudiation and allows verification independent of transport security.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-18T21:11:35.132290+00:00— report_created — created