Concepts
Deep dives into the theory and practice of AI agent governance.
Core Concepts
What is Agent Governance?
The foundational guide to understanding why AI agents need governance. Covers the four pillars: Identity, Policy, Enforcement, and Audit.
Trust Tiers Explained
How MeshGuard uses graduated trust levels to enable different autonomy levels for different agents. From anonymous to privileged.
Principle of Least Privilege
Applying the security principle of least privilege to AI agents. Grant only the permissions agents need, nothing more.
Delegation Chains
How agents can delegate permissions to other agents in a controlled, auditable way. Covers scope restrictions, depth limits, and revocation.
Policy Design Patterns
Best practices for writing MeshGuard policies that scale. Covers deny-by-default, role-based, time-based, and resource-based patterns with real-world examples.
Audit Logs for Compliance
How MeshGuard's audit logging supports compliance requirements. Covers log structure, retention, querying, and export for SOC2, HIPAA, and GDPR.
Trust & Behavior
Behavioral Trust Scoring
How multi-component behavioral scoring computes dynamic trust for AI agents. Covers weighted linear combination, component independence, normalization, and tier quantization.
Anomaly Detection in Agent Meshes
Detecting anomalous agent behavior through constraint violation over trust graph structures. Severity scoring, auto-response feedback loops, and real-time vs. batch detection.
Delegation Chains Deep Dive
A rigorous exploration of delegation chain mechanics: DAG enforcement, scope monotonicity, depth bounding, time-bounded permissions, and the relationship to capability-based security.
Trust Graph Architecture
The flagship technical paper on modeling agent governance as a directed acyclic graph. Formal definitions, behavioral trust computation, anomaly detection, delegation protocols, and security analysis against compromised-agent and collusion threat models.
Learn by Doing
Chat with Scout, our governed demo agent, to see these concepts in action.