Agent Communication Protocol Security

Agent communication protocols define how agents exchange messages, invoke tools, and share state. The security properties of these protocols determine whether inter-agent communication can be trusted. A protocol that lacks authentication, encryption, or integrity checking is an open door for attackers.

Protocol Threat Model

Four threats apply to any agent communication protocol:

Eavesdropping: An attacker reads messages between agents, extracting sensitive data or learning about system behavior.

Tampering: An attacker modifies messages in transit, changing action parameters, injecting instructions, or corrupting data.

Spoofing: An attacker sends forged messages that appear to come from a legitimate agent.

Replay: An attacker captures a legitimate message and resends it later to trigger duplicate actions.

Transport Security

All inter-agent communication should use encrypted transport. For agents communicating over a network, TLS is the baseline requirement. For agents in the same process, shared memory channels should be access-controlled to prevent unauthorized reads.

Message Format Security

The message format should include fields that support security verification: sender identity, recipient identity, timestamp, nonce, and cryptographic signature. The Model Context Protocol (MCP) provides structured tool invocation but does not natively include all these fields. Authensor's envelope format adds the missing security fields on top of MCP.

Schema Validation

Every message should be validated against a strict schema before processing. Reject messages with unexpected fields, missing required fields, or fields that do not match expected types. Schema validation prevents many injection attacks that rely on smuggling unexpected content into message fields.

Rate Limiting

Apply rate limits at the protocol level. An agent that suddenly sends thousands of messages per second is likely compromised or malfunctioning. Protocol-level rate limiting prevents a single agent from overwhelming the communication infrastructure.

Secure communication is not a feature to add later. It is a foundational requirement that must be present from the first inter-agent message.