ai-safe2-framework

Incident evidence and accountability review

Reconstruct activity, preserve evidence, and assign follow-up authority

AI SAFE² Guide

Framework Home Cross-Pillar Governance AISM NEXUS Technology Profile

Evidence workflow

This assessment guidance strengthens review of logging, traceability, incident response, oversight, and learning under existing controls. It does not add statutory deadlines, determine legal liability, or establish that a notified organization was breached. The named legal/incident owner determines applicable retention, legal hold, privacy, and external notification obligations.

  1. Reconstruct: Preserve system/version identity, trusted request and approvals, grants/delegation, tool calls and arguments, policy decisions, side effects, shutdown events, timestamps, and external observations. Identify collection gaps.
  2. Identify affected parties: Separate attempted access, public retrieval, suspected impact, and confirmed unauthorized state changes. Retain evidence for each classification and mark unknown target impact explicitly.
  3. Preserve: Assign a custodian, restricted access, retention schedule, integrity/provenance records, and a documented hold decision. Minimize sensitive data and retain originals safely; do not confuse a digest with authenticated truth.
  4. Notify through the owner: Record recipients, basis, facts/uncertainties, approval, channel, timestamp, correction, and receipt where appropriate. A report or repository recommendation is not authorization to contact anyone.
  5. Learn and recover: Connect the finding to controls, containment/recovery, policy changes, and an isolated regression test with an owner and review date.

Control pointers

Purpose Existing control references Repository guidance
Reconstruction P2.T3.1 Real-Time Activity Logging; P2.T3.7 Decision Traceability Ledger
Containment and response P3.T5.1 Circuit Breakers; P3.T5.2 Emergency Shutdown; P3.T5.10 Incident Response Circuit Breaker
Detection and records P4.T8.2 Anomaly Detection; P4.T8.3 Security Logging Command Center
Learning P5.T9.1 Threat Intel Integration; P5.T9.2 Playbook Updates; CP.6 AI Incident Feedback Loop Integration Learning Engine, Cross-Pillar Governance
Decision authority CP.10 HEAR Doctrine (Human Ethical Agent of Record) HEAR doctrine

Use the Technology Card to distinguish an incident source that evidences or challenges a control from a product that enforces it. Accountability events belong in governance/lifecycle review; they cannot be credited as technical validation. The example backlog preserves source limitations and pending verification for the supplied events.


Evidence Assurance Validation Security Policy

AI SAFE² v3.1 · Cyber Strategy Institute