liveBook · Manning

Multi-agent governance

As explained in AI Governance

Multi-agent governance is the definition and enforcement of rules for identities, permissions, communication, and failure handling among multiple agents.

When agents communicate, governance defines who they trust and what they may ask one another to do. Without explicit controls, a malicious or fake agent could impersonate a legitimate one to access data or issue commands. section 4.8.8

Set controls for inter-agent trust and compromise:

  • Specify which agents may communicate and what requests they may make. section 4.8.8
  • Require mutual authentication and monitor communications for anomalies. section 4.8.8
  • Plan to revoke a compromised agent’s identity and isolate it from the system. section 4.8.8

How AI Governance covers it

How deep it goes
Specifies trust, authentication, monitoring, and containment controls without detailing their implementation.
Examples
Illustrates impersonation risk with a malicious or fake agent posing as a legitimate one.
What it assumes
Assumes multiple agents communicate.
What you can do after
Configure permitted agent interactions, mutual authentication, anomaly monitoring, and compromise containment.
How it connects
Connects agent trust relationships with mutual authentication, communication monitoring, identity revocation, and isolation.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of AI Governance. The section links point to the parts of the book it draws on.