"Insecure Inter-Agent Communication occurs when the channels, protocols, or message formats used by agents to communicate with each other lack sufficient authentication, integrity verification, or access controls — allowing attackers to forge, intercept, replay, or manipulate messages to cause agents to perform unauthorized actions."
Authentication answers: is this message really from the agent it claims to be from? Without authentication, any party that can write to the channel — or knows the message format — can send a message that an agent will treat as legitimate. In many current multi-agent implementations, agents authenticate to external APIs using credentials but apply no authentication between agents in the same pipeline. Internal trust is assumed, not verified.
Integrity verification answers: is this message the same one that was sent? Without integrity checks (cryptographic signatures, MACs, checksums), a message can be modified between send and receive without detection. Authentication alone isn't enough — a signed envelope can contain a tampered letter if only the envelope is verified.
The attacker's goal is to make an agent do something it wouldn't do if it knew the true origin or content of the message. Forging = creating a new fake message. Intercepting = reading messages in transit. Replaying = re-sending a previously valid message at an unauthorized time. Manipulating = altering a message in flight. All four exploit the same underlying gap: the receiving agent has no way to verify what it received.