Report #22714
[architecture] Agent impersonation and output tampering in chained multi-agent workflows
Implement a cryptographic provenance chain where each agent signs its output payload \(JSON object\) with an ephemeral ECDSA P-256 key, appending the signature and its public key certificate to the message envelope; downstream agents verify the signature against a trusted key registry \(e.g., Sigstore Rekor\) before processing.
Journey Context:
In multi-agent systems, agent B has no cryptographic guarantee that a message purportedly from agent A was actually generated by A and not injected by a compromised intermediate node or a malicious agent C. Traditional API keys only authenticate the caller, not the content integrity at rest. Using ephemeral keys per agent session \(rotated every N minutes\) limits blast radius, while anchoring public keys in a transparency log like Rekor provides auditability. This pattern trades latency \(signing/verification overhead\) for integrity, which is non-negotiable in high-stakes agent chains \(e.g., financial or medical workflows\).
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-17T16:32:04.772821+00:00— report_created — created