Security

Overview

AgentTrust implements defense-in-depth security: authentication, encryption at rest, audit integrity, PII detection, and tier-based authorization. This page summarizes security controls for operators and developers.

Why It Matters

AI governance systems handle sensitive agent inputs/outputs. Security misconfiguration can expose audit data or bypass policy enforcement.

Prerequisites

• Full edge gateway for production security features
• Enterprise tier for SSO, hash chain, SOC2 export

Step-by-Step Guide

Authentication

Critical: Set AUTH_ENABLED=true in all non-dev environments.

Authorization

• JWT contains tier claim — gateway enforces capability matrix
• Rate limiting: 60 req/min per agent_id (configurable)
• Team management via /v1/team/*

Encryption

Audit integrity

• Append-only ledger with content hashes
• Hash chain verification: GET /v1/audit/chain/verify (Enterprise)
• Ledger hashes are permanent (required for integrity; not erasable)

PII handling

• Policy engine detects SSN, email, API keys, passwords
• PII fields masked in hot store within 24 hours
• Erasure endpoint: DELETE /v1/audit/executions/{id}/pii
• LLM judge receives truncated input/output only (never full payload)

Secret management

Kill-switch

export AGENTRUST_ENABLED=false

Instant rollback — all governance paths become no-ops.

Examples

Verify auth:

curl -H "X-AgentTrust-Token: $AGENTRUST_KEY" http://localhost:8000/v1/auth/check

Best Practices

• Never commit API keys or JWT secrets to version control
• Use secret managers (Vault, AWS Secrets Manager, K8s Secrets)
• Enable audit encryption in production
• Run AUTH_ENABLED=true always in production
• Complete AIIA for high-risk domains
• Follow Incident Response procedures

Common Mistakes

• AUTH_ENABLED=false in production
• Exposing gateway port 8000 without TLS/reverse proxy
• Storing production keys in .env files in git repos
• Using embedded gateway for regulated production data

Troubleshooting

Related Resources

• Data Governance
• Compliance
• Incident Response
• .github/SECURITY.md