AG07:2026 — Insecure Inter-Agent Communication

AG07:2026 Insecure Inter-Agent Communication

Slide 25 of 28

Slide 25 · Quiz — Questions 3, 4 & 5

3. An agent pipeline operates entirely on an internal corporate network with no external internet access. A security architect argues that TLS between agents is unnecessary because the traffic never leaves the internal network. What is wrong with this reasoning?

Nothing — internal-only traffic does not require encryption, so the architect is correct Internal networks are not inherently trusted: compromised internal services, overpermissioned logging tools, and shared cloud VPC segments can all read unencrypted internal traffic TLS is required only when agents communicate with external APIs — internal agent-to-agent traffic is out of scope for TLS TLS adds too much latency for internal agent communication and should be replaced with application-layer encryption instead

4. What makes orchestrator impersonation more dangerous than impersonating a specialized sub-agent (such as a data retrieval agent)?

Orchestrators are harder to compromise because they run on more secure infrastructure Sub-agents verify the orchestrator's identity more rigorously than the orchestrator verifies sub-agent identities Impersonating the orchestrator gives the attacker authority over all downstream sub-agents simultaneously, because sub-agents are designed to follow orchestrator instructions Orchestrators use stronger cryptographic keys than sub-agents, making impersonation harder to achieve

5. A team is building a new agent pipeline and must choose the minimum viable set of mitigations for a first deployment. Which combination addresses the broadest set of attack types for the lowest implementation cost?

MIT08 (audit logging) + MIT09 (anomaly detection): detect attacks quickly and respond MIT04 (input validation) + MIT05 (replay protection): validate every message and ensure freshness MIT01 (agent authentication) + MIT03 (mTLS) + MIT02 (message signing) + MIT05 (replay protection): authenticate identities, encrypt channels, sign content, and prevent replays MIT07 (agent registry) alone: a central registry of trusted agents addresses all inter-agent trust issues

← Back Complete AG07 →