The five VERITRAIL pillars divide AI governance into parts that each have a clear owner, a clear purpose and clear evidence. They are deliberately ordered: accountability comes first because without decision rights nothing else sticks, and assurance comes last because it tests whether everything before it actually works.
| Pillar | Prefix | Controls | Core question | Typical accountable owner |
|---|---|---|---|---|
| Accountability & Operating Model | AO | 12 | Who owns AI risk, who decides, and how is governance resourced? | Executive sponsor, AI Governance Lead |
| Enterprise AI Inventory | EI | 12 | What AI do we run, who owns it, and how risky is it? | AI Governance Lead, CIO office |
| Governance & Lifecycle Control | GL | 12 | What gates does an AI system pass from intake to retirement? | AI CoE, AI Risk Lead, Governance Forum |
| Integrity, Safety & Security | IS | 16 | Are data, fairness, safety, security and runtime behavior controlled? | CISO, Privacy Lead, Platform Engineering |
| Stewardship, Evidence & Assurance | SE | 12 | Can we prove it, learn from incidents and improve? | AI Governance Lead, second line, Internal Audit |
How the pillars work together
Consider a business team that wants to deploy an agent that drafts customer refund decisions and can issue refunds through a payments API.
- Accountability (AO). The use case is screened against the prohibited and restricted use policy (AO-10). The risk appetite (AO-04) determines which tier triggers forum review (AO-02).
- Inventory (EI). The use case enters the intake register (EI-12). The agent, its model, its prompts and the payments tool are registered (EI-05, EI-07) with a named business and technical owner (EI-03) and a provisional risk tier that a human confirms (EI-02).
- Lifecycle (GL). Triage routes it to a full pre-deployment assessment (GL-01, GL-02). Approval is granted with conditions (GL-03), and go-live is blocked until those conditions are verified (GL-07).
- Integrity, Safety & Security (IS). The agent gets only the tools it needs (IS-11). Refunds above a threshold require human approval at runtime (IS-13), customer data is redacted before external model calls (IS-04), and the agent is tested for prompt injection before release (IS-14). A kill switch is tested (IS-15).
- Assurance (SE). Every policy decision, tool call and approval is logged tamper-evidently (SE-08). Monitoring and drift thresholds feed reviews (SE-12), incidents follow a defined procedure (SE-04), and an evidence pack can be produced on request (SE-05).
No single pillar is sufficient. An inventory without lifecycle gates is a list. Runtime guardrails without accountability are configuration nobody owns. Assurance without the first four pillars has nothing to test.
Pillars and the VERITRAIL modules
| VERITRAIL module | Primary pillars | What it operationalizes |
|---|---|---|
| Assess | AO, EI, GL | Maturity baseline, use-case triage and risk/impact assessment |
| Register | EI | AI system of record, agent and tool registry, vendor AI inventory |
| Control | AO, GL | Policies, approval gates, change triggers, exceptions |
| Shield | IS | Runtime policy, redaction, least privilege, human approval, kill switch |
| Assurance | SE | Control testing, logging, incidents, evidence packs, monitoring |
| Agent | All | Citation-backed drafting of assessments, gap registers and reports |
Where to begin
If you are unsure where you stand, take the maturity self-assessment. It scores eight domains that cut across the pillars and flags critical gaps — for example, no reliable inventory (EI-01, EI-09) or no tested kill switch (IS-15) — so you can prioritize the right pillar first.
VERITRAIL controls cited
- AO-02 Executive accountability and AI governance forumNIST AI RMF: GOVERN 2.1, GOVERN 2.3ISO/IEC 42001: 5.1, 5.3, A.3.2EU AI Act: Art. 17
- EI-01 Enterprise AI inventory (system of record)NIST AI RMF: GOVERN 1.6ISO/IEC 42001: A.4.2, 7.5EU AI Act: Art. 26, Art. 49
- GL-02 Pre-deployment risk and impact assessmentNIST AI RMF: MAP 1.1, MAP 3.1, MAP 5.1, MEASURE 1.1ISO/IEC 42001: 6.1.4, 8.4, A.5.2, A.5.3, A.5.4EU AI Act: Art. 9, Art. 27
- IS-13 Human approval for high-impact actionsNIST AI RMF: GOVERN 3.2, MANAGE 2.2ISO/IEC 42001: A.9.2, A.6.2.6EU AI Act: Art. 14
- SE-08 Immutable action loggingNIST AI RMF: MEASURE 2.8, MANAGE 4.1ISO/IEC 42001: A.6.2.8EU AI Act: Art. 12, Art. 19, Art. 26
Framework mappings are indicative VERITRAIL interpretations (Control Library v1.0.0); they are not legal advice and do not establish conformity or compliance.
The complete list of controls, with objectives, evidence and indicative mappings, is in the VERITRAIL Control Library.