Chapter 6: Audit & Compliance

Outline

6.1 Introduction to Audit and Compliance

• Why agent systems need evidence, not just logs
• OpenClaw’s audit goals: accountability, forensics, operator trust, and regulatory proof
• The difference between runtime observability and defensible evidence

6.2 Event Collection at Security Boundaries

• Capturing events from ingress, policy, execution, tools, approvals, outbound sends, and system state
• Logging decisions as well as actions
• Why missing boundary events make later investigation unreliable

6.3 Normalization, Enrichment, and Correlation

• Converting heterogeneous events into a consistent schema
• Preserving provenance, session context, actor identity, and policy references
• Correlating multi-step workflows without inventing certainty

6.4 Evidence Integrity, Retention, and Access Control

• Append-only evidence stores, hash chaining, retention classes, and archival
• Separating operational search from long-term preservation
• Protecting audit access from becoming a new disclosure risk

6.5 Identity, Policy, and Approval Auditing

• Recording authentication, authorization, capability grants, denials, and step-up checks
• Auditing human approval as a first-class security event
• Proving who authorized a high-impact action and under what scope

6.6 Compliance Mapping and Evidence Packages

• Mapping OpenClaw events to control families without turning the platform into checkbox theater
• Building evidence packs for SOC 2, ISO 27001, NIST-oriented reviews, and internal audits
• Showing control operation over time instead of relying on screenshots

6.7 Privacy-Aware Logging and Sensitive Data Handling

• Data minimization, selective redaction, and content-free decision records
• Logging enough to investigate without creating a shadow data lake of secrets
• Handling messages, attachments, and provider responses safely

6.8 Operating the Audit Pipeline

• Backpressure, health checks, loss detection, and clock discipline
• Monitoring the audit system itself
• What to do when evidence capture is degraded

6.9 Incident Response and Forensic Reconstruction

• Rebuilding the chain from user intent to tool execution to external outcome
• Distinguishing proven facts, inferred relationships, and unknowns
• Preserving evidence while containing the incident

6.10 Third-Party, Supply Chain, and Administrative Evidence

• Provider calls, dependency updates, and administrative changes as audit subjects
• Tracking configuration drift and policy changes
• Keeping vendor and platform dependencies visible in the evidence story

6.11 Case Study: Investigating a Disputed Outbound Send

• A realistic OpenClaw incident involving a contested external message
• How the audit pipeline answers what happened, what was approved, and what was blocked
• Lessons for operators and reviewers

6.12 Chapter Summary

• Design principles and implementation checklist
• Common audit anti-patterns to avoid
• Preparing for deployment and incident chapters that build on this evidence model