AI 扩展计划 / Playbooks

AI Three Lines Governance / Decision Rights / Assurance Playbook

版本: v1.0

459 行AI_THREE_LINES_GOVERNANCE_DECISION_RIGHTS_ASSURANCE_PLAYBOOK.md

AI Three Lines Governance / Decision Rights / Assurance Operating Model Playbook

版本: v1.0 日期: 2026-06-30 适用对象: experienced CBAP、financial retail PM、product architect、enterprise architect、AI governance lead、risk / compliance partner、internal audit partner、AI platform owner。

本文是一份执行型手册, 用于把 AI use case 的三道防线职责、生命周期 gate、decision rights、evidence forum、challenge memo、issue/action tracking、assurance packet、escalation path 和 architecture governance 连接成可运行的 operating model。本文不构成法律意见、监管解释、审计意见、模型验证结论、内控有效性结论或生产批准。

1. Purpose and When To Use

Use this playbook when an AI initiative has any of these conditions:

Trigger	Why this playbook is needed
AI output affects customers, cases, credit, onboarding, financial crime, complaints or regulated content	Decision rights and control ownership must be explicit before release.
RAG, copilots, agents or vendor models are embedded into business workflow	Evidence must connect architecture, controls, issues and gate decisions.
Risk, compliance, privacy, security, model risk, third-party risk or internal audit will review the use case	First-line ownership, second-line challenge and third-line assurance must not be blurred.
Project teams say "risk signed off" without naming residual-risk owner	The operating model needs a clear decision record and challenge memo.
Conditional release or management action is expected	Issue/action tracking needs owner, due date, acceptance criteria, closure evidence and escalation path.
Similar AI issues repeat across teams	Architecture governance should turn repeat issues into reusable controls and platform standards.

Recommended use cases:

Use case	Operating-model focus
GenAI contact center	Regulated content boundary, escalation, quality review, transcript evidence
AML copilot	Analyst-owned disposition, source-span evidence, typology coverage, case narrative review
Credit decision support	Decision-support boundary, policy consistency, adverse-action evidence, fairness challenge route
KYC onboarding	Document classification, missing-information follow-up, false reject / false accept issue route
AI vendor model	Data use terms, model update notice, incident notice, fallback and exit readiness
Enterprise knowledge assistant	Approved-source governance, entitlement-aware retrieval, issue routing for stale or sensitive content

Definition:

AI Three Lines operating model = first-line product/ops ownership
  + second-line independent challenge
  + third-line independent assurance
  + lifecycle decision rights
  + evidence forum
  + issue/action tracking
  + architecture governance integration

2. Operating Model

2.1 Core Principles

Principle	Practical rule
First line owns product and operational risk	Product / operations owner owns outcome, control operation, evidence production and residual-risk proposal.
Second line challenges, conditions and escalates	Risk / compliance / privacy / security / model risk / TPRM define standards and challenge evidence.
Third line assures independently	Internal audit does not approve releases; it provides independent assurance through audit planning, testing and findings.
Gate decisions bind exact evidence	Every gate decision references evidence packet version, issue disposition and conditions.
Conditions are managed actions	Conditional go decisions require owner, due date, acceptance criteria and closure evidence.
Architecture governance is part of the model	ADRs, ARB minutes, reference architecture and platform standards must reflect control and evidence decisions.
Material change reopens review	Model, prompt, RAG corpus, tool permission, automation, vendor or customer-impacting workflow changes trigger gate analysis.

2.2 Operating Cadence

Cadence	Purpose	Participants	Output
Intake triage	Classify AI use case and risk tier	PM, governance office, EA, relevant second line	use case record, gate plan
Design evidence review	Challenge architecture, controls and evidence plan before pilot	Product architect, control owners, risk/compliance, platform	design decision, issue log
Release evidence forum	Review release packet, challenge memo and open issue disposition	First line owner, second line challenge owners, governance lead, EA	gate decision record
Issue/action review	Track conditions, defects, overdue actions and repeat issues	action owners, governance office, second line	action status, escalation
Architecture governance review	Convert repeat risks into standards and reusable platform patterns	EA, platform, security, governance office	ADR, reference architecture update
Assurance planning touchpoint	Align audit universe with high-risk AI areas without making audit an approver	internal audit, governance office, relevant management	assurance plan inputs

3. Template: Three-Lines Role Map

Use this table at intake and update it at each material gate.

Role area	First line product / ops	Second line risk / compliance	Third line internal audit	Evidence
Accountable owner	Business sponsor: Head of Financial Crime Operations; product owner: AML Copilot Product Owner; operations owner: AML Investigation Manager	Challenge functions: Financial Crime Compliance, Model Risk, Privacy, Security, Third-Party Risk	Audit relationship owner: Internal Audit Financial Crime Technology Director	owner attestation, AI inventory record
Business outcome	Defines benefit, scope, users, process change and operating capacity	Challenges whether outcome creates unmanaged risk or policy conflict	May assess whether governance representations are reliable	PRD, value case, operating model
Risk tier and scope	Proposes risk tier and automation boundary	Challenges risk tier, customer impact and policy mapping	May use risk tier for audit universe planning	risk tier worksheet
Control ownership	Owns control design and operating performance	Challenges control adequacy and failure condition	Tests selected controls if in audit scope	control matrix
Evidence ownership	Produces evidence packet and maintains evidence freshness	Challenges completeness, relevance and expiry	Uses evidence for independent assurance	evidence packet index
Gate decision	Makes go / no-go / conditional-go recommendation within authority	No objection, condition, object or escalate	No release approval; assurance note if relevant	gate decision record
Issue closure	Owns management actions and closure evidence	Challenges severity, due date and closure sufficiency	Validates closure for audit findings where applicable	issue/action log
Architecture governance	Implements approved architecture and control points	Challenges policy and risk fit of architecture	Reviews governance consistency if audited	ADR, ARB minutes

Completed example:

Role area	GenAI contact center example
First line	Contact center director owns agent-assist operating process; AI product owner owns release packet; quality operations owns sampling control.
Second line	Compliance challenges regulated answer boundaries; privacy challenges transcript retention; conduct risk challenges vulnerable customer escalation.
Third line	Internal audit does not approve launch; it records the use case for possible future review of governance and control operation.

4. Template: Lifecycle Decision-Rights Matrix

Gate	Decision question	First line decision right	Second line challenge right	Third line assurance role	Required evidence	Possible decision
Intake	Should this be registered and assessed as AI?	Accept use case scope and owner	Challenge AI classification, risk tier and policy functions in scope	None or audit universe awareness	use case record, risk tier worksheet	proceed, reclassify, reject
Design	Is design fit for controlled pilot?	Approve design proposal and control owner map	Challenge control design, architecture, data use and evidence plan	Review artifacts only if assurance planning requires	architecture view, control matrix, evidence plan	proceed, redesign, escalate
Pilot	Is limited pilot controlled enough?	Approve pilot scope, users, traffic and exit criteria	Challenge pilot population, restrictions, monitoring and stop rule	May observe governance, not approve pilot	pilot plan, eval summary, SOP, issue log	pilot, reduce scope, no-go
Release	Can production release occur?	Make go / no-go recommendation and accept residual-risk proposal	No objection, condition, objection or escalation	Does not sign release; may reference prior findings	release packet, challenge memo, open issue disposition	go, conditional go, no-go, escalate
Scale	Can traffic, automation or scope expand?	Request scale based on value and control evidence	Challenge evidence under expanded population and capacity	May include in audit plan	scale memo, operating evidence, action aging	scale, hold, reduce, rework
Material change	Can changed model, prompt, RAG, tool or vendor proceed?	Classify change and request approval	Challenge materiality and regression evidence	May test change process if audited	change impact, regression eval, ADR update	approve, require gate, rollback
Incident / stop	Should system pause, degrade or rollback?	Execute stop rule and remediation	Challenge severity, customer impact and closure	Independently review incident handling later	incident record, replay packet, action plan	pause, rollback, continue with controls
Retire	Can capability or dependency be decommissioned?	Retire product scope and operational dependency	Challenge retention, obligations and customer impact	May test retirement controls	retirement ADR, retention evidence	retire, extend support, remediate

Decision vocabulary:

Decision	Meaning
Go	Evidence sufficient, issues within authority, controls ready
Conditional go	Release allowed only with named actions, due dates and review trigger
No-go	Blocker issue or evidence gap prevents release
Reduce scope	Pilot or release continues with lower-risk users, channels, actions or data
Escalate	Authority exceeded or first / second line disagreement requires senior decision
Pause / rollback	Production control failure, incident or material drift requires stop action
Retire	Use case, model or vendor dependency exits with retention and obligation evidence

5. Template: Evidence Forum Agenda

Use for design, release, scale and material-change gates.

Agenda item	Owner	Evidence	Decision needed
1. Confirm gate and decision authority	Governance lead	decision-rights matrix	Confirm who can decide, challenge and escalate
2. Confirm use case scope and changes	PM	use case record, change note	Confirm scope, automation boundary and customer impact
3. Review architecture and control points	Product architect / EA	architecture diagram, ADR	Confirm policy decision point, evidence capture and issue route
4. Review control ownership	First line control owners	control matrix	Confirm owner, frequency, failure condition and evidence
5. Review eval / test / pilot evidence	AI platform / delivery lead	eval report, pilot findings	Confirm release criteria and failure disposition
6. Review second-line challenge memo	Challenge owners	challenge memo	Confirm no objection, conditions, objections or escalation
7. Review issue/action log	Governance lead	issue register	Classify blockers, conditions, accepted issues and overdue actions
8. Decide gate outcome	Accountable first line owner	evidence packet	Record go, conditional go, no-go, reduce scope or escalate
9. Confirm ADR and review triggers	Architect	ADR, review triggers	Update architecture decision and material-change triggers
10. Confirm management action follow-up	Action owners	action tracker	Assign closure evidence and next review date

Forum discipline:

Do not review every artifact page by page; use evidence index and exception-based review.
Do not allow verbal closure of issues; closure requires acceptance criteria and evidence.
Do not combine risk acceptance with unclear owner language; record first-line accountable owner.
Do not treat internal audit attendance as release approval.

6. Template: Second-Line Challenge Memo

Concrete memo example:

# Second-Line Challenge Memo: AML Copilot Release Gate

Date: 2026-06-30
Gate: release
Challenge functions: Financial Crime Compliance, Model Risk, Privacy, Security
First-line accountable owner: Financial Crime Operations Director

## 1. Scope Reviewed

- Use case: AML copilot drafts sourced investigation timelines and narrative suggestions for alert types A and B.
- Channel / user population: L2 AML analysts in the retail banking investigations queue.
- Automation boundary: The copilot cannot close alerts, change disposition, file SARs or send customer communications.
- Customer / employee / regulatory impact: Analyst productivity and case narrative quality; regulatory sensitivity through case record content.
- Model / RAG / agent / vendor components: Enterprise LLM gateway, approved AML procedure corpus, case timeline retriever, narrative drafting copilot.
- Evidence packet version: AML-COPILOT-REL-2026-06-30-v1.

## 2. Challenge Summary

| Area | Challenge position | Required condition or rationale |
|---|---|---|
| Risk tier and scope | conditional no objection | High-risk operational support is acceptable for pilot-to-production only because final disposition remains analyst-owned. |
| Control design | condition | Narrative evidence must bind each saved paragraph to source-span ids and analyst approval event. |
| Evidence sufficiency | condition | Current eval report is sufficient for common typologies but needs 30-case source-span sample before broad scale. |
| Operational readiness | no objection | SOP, reviewer training and fallback to manual narrative drafting are documented. |
| Change and incident readiness | condition | Material changes to AML corpus, prompt or narrative template must reopen release gate. |

## 3. Material Concerns

| Concern id | Concern | Severity | Gate impact | Suggested action |
|---|---|---|---|---|
| C-001 | Paragraph-level source-span evidence is incomplete for high-risk typology narratives. | high | conditional release only | Add event capture and produce 30-case source-span review sample. |
| C-002 | Material-change trigger for AML corpus updates is not reflected in ADR. | medium | condition | Update ADR with corpus, prompt and workflow materiality rules. |

## 4. Conditions for Proceeding

| Condition id | Condition | Owner | Due date | Closure evidence |
|---|---|---|---|---|
| CON-001 | Capture source ids, passage ids, analyst edit diff and approval event for every narrative saved to case record. | AML Copilot Product Owner | 2026-07-15 | Event extract, 30-case sample workbook, reviewer sign-off. |
| CON-002 | Update ADR with material-change triggers for corpus, prompt, workflow and vendor gateway changes. | Product Architect | 2026-07-10 | Approved ADR-AML-COPILOT-004. |

## 5. Residual Concerns and Review Triggers

- Residual concern: Analysts may over-rely on fluent narrative drafts despite source support.
- Review trigger: Source-span error above 2% in monthly sample, any unsupported SAR-related phrase, or material corpus update.
- Escalation trigger: Any production narrative saved without source-span event.

## 6. Challenge Position

Final position: conditional no objection for controlled production release, no broad scale until CON-001 and CON-002 close.

Summary view:

Field	AML copilot release example
Concern	Draft investigation narrative lacks paragraph-level source-span evidence for high-risk typologies.
Severity	High
Gate impact	Conditional go for pilot only; production release blocked until evidence is available.
Condition	Every saved narrative must capture source ids, analyst edit diff and approval event for a 30-case sample.
Closure evidence	Event extract, sample review workbook and updated ADR showing evidence collector design.

7. Template: Issue / Action Log

Field	Description	Example
Issue id	Unique issue identifier	ISS-KYC-2026-014
Use case / gate	AI use case and lifecycle gate	KYC onboarding / pilot
Issue type	evidence gap, control design gap, operation failure, policy deviation, unresolved challenge, issue aging, repeat finding	control design gap
Issue statement	Specific problem, not vague risk language	Missing-info email draft can be sent without reviewer approval in edge case path
Severity	critical, high, medium, low	high
Gate impact	blocker, condition, monitor, accepted risk, escalate	blocker
Root cause hypothesis	Why it exists	workflow branch bypassed approval state
Action statement	Concrete management action	Add approval state check before send event and negative test for all branches
Action owner	Named role	KYC product owner
Due date	Date	2026-07-10
Acceptance criteria	Testable closure rule	100% negative tests block unapproved send; sample trace shows approval event before send
Closure evidence	Evidence required	test report, trace sample, ADR update
Second-line view	challenge / closure opinion	Compliance confirms approval path is sufficient
Escalation trigger	When to escalate	overdue by 7 days or any production bypass
Status	open, in progress, ready for review, closed, reopened, escalated	in progress

Issue severity guide:

Severity	Definition	Typical action
Critical	Customer harm, regulatory breach, unauthorized write action or severe evidence failure likely or observed	stop, rollback, executive escalation
High	Material control gap or unresolved second-line objection for high-risk use case	release blocker or conditional pilot only
Medium	Control or evidence gap with compensating control and limited exposure	conditional release with due date
Low	Documentation, clarity or low-risk evidence improvement	track to next gate

8. Template: Assurance Packet

Assurance packet is the minimum set of artifacts that lets reviewers understand what was decided, on what evidence, by whom, with which issues and review triggers.

Section	Required content	Owner
Cover page	use case id, gate, date, decision, accountable owner, evidence packet version	Governance lead
Scope and boundary	business objective, user group, data, model, RAG, agent, tool, vendor, automation boundary	PM / architect
Three-lines role map	first-line owners, second-line challenge owners, third-line assurance role	Governance lead
Architecture view	context diagram, policy decision points, evidence collector, issue route, fallback	Product architect / EA
Control matrix	control objective, activity, owner, evidence, frequency, failure condition	Control owners
Eval / test summary	test scope, thresholds, failures, defect disposition, unresolved limits	AI platform / delivery
Challenge memo	second-line position, objections, conditions, residual concerns	Second line owners
Issue/action log	blocker, condition, accepted issue, action status and aging	Governance lead
Gate decision record	decision, rationale, conditions, review triggers, escalation path	First line owner + governance lead
ADR	architecture decision, tradeoffs, consequences, review triggers	Architect
Operating evidence plan	post-release evidence cadence, issue review cadence, scale criteria	Operations owner

Quality bar:

Every claim in executive narrative links to evidence.
Every material issue has owner, due date and acceptance criteria.
Every conditional go has explicit follow-up forum or review date.
Every architecture tradeoff appears in an ADR.
Every high-risk action has policy and evidence route.

9. Template: Escalation Path

Trigger	First response	Escalation owner	Decision authority	Evidence required
Release blocker unresolved	Hold gate or reduce scope	Governance lead	Business sponsor + risk acceptance authority	blocker issue, impact, options
First / second line disagreement	Document positions and options	Governance chair	Delegated executive / risk committee	decision memo, challenge memo
Conditional action overdue	Notify owner and sponsor	Governance office	Portfolio governance / sponsor	action aging report
Production control failure	Pause affected function or activate fallback	Incident commander	Incident management authority	incident record, trace, customer impact
Repeat issue across use cases	Create platform standard candidate	Enterprise architect	Architecture review board	trend, root cause, proposed standard
Vendor material change	Freeze change or restrict use	Vendor owner / TPRM	Third-party risk authority + sponsor	vendor notice, impact analysis
Audit finding overdue	Follow audit issue route	Management action owner	Audit issue governance	action status, closure evidence

Escalation decision memo:

Field	Content
Decision required	go, no-go, reduce scope, accept risk, extend due date, pause, rollback
Options	option A / B / C with benefit, risk and operational impact
First-line position	accountable owner recommendation
Second-line position	no objection, condition, objection or escalation rationale
Assurance consideration	relevant audit finding or assurance concern, if any
Evidence	packet ids, issue ids, ADR ids
Final decision	selected option, owner, expiry, review trigger

10. PM / BA / Architecture Questions

PM Questions

Question	Strong answer should include
Who owns the business outcome and residual-risk proposal?	Named first-line owner and delegated decision authority
Which lifecycle gate are we at?	Intake, design, pilot, release, scale, change, incident or retire
What is the AI allowed and not allowed to do?	Automation boundary, prohibited actions and human workflow
What evidence proves value and control readiness together?	KPI, KRI, KCI and evidence packet links
What happens if second line objects?	Escalation path and decision authority
What conditions can we accept without weakening control?	Action owner, due date, acceptance criteria and review trigger

BA Questions

Question	Strong answer should include
What claims must be verifiable at the gate?	Claim-to-control-to-evidence mapping
Which process states create control evidence?	Workflow state model, event dictionary and evidence owner
Which exceptions are allowed and how are they closed?	Exception taxonomy, owner, expiry and compensating control
How is issue severity determined?	Gate impact, customer impact, policy deviation and evidence failure criteria
What is the minimum sufficient evidence packet?	Artifact list with owners and version ids
How are management actions tested before closure?	Acceptance criteria and closure evidence

Architecture Questions

Question	Strong answer should include
Where are policy decisions enforced?	Policy decision point, enforcement point and decision log
Where is evidence generated by design?	Trace store, evidence collector, event schema and retention tags
How do RAG / agent / copilot components map to controls?	Source inventory, tool registry, approval binding and UX state
Which architecture changes reopen the gate?	Model, prompt, RAG corpus, tool permission, vendor, automation and workflow materiality
How do issues become platform improvements?	Repeat issue trend, reference architecture standard and ADR
How can audit or independent reviewers query evidence without joining delivery tools manually?	Evidence packet index, access control and query path

11. Release Checklist

Use this as a gate-entry checklist. A release meeting should not begin until the required items exist.

Check	Pass criteria	Gate-entry signal
Use case registered	AI inventory record has owner, scope, risk tier and automation boundary	Missing record defers forum
Three-lines role map complete	First-line owner, second-line challenge owners and third-line role are documented	Missing owner blocks gate
Decision authority confirmed	Gate decision owner and escalation authority are known	Missing authority escalates before forum
Architecture view complete	RAG / agent / copilot / policy / evidence / issue route are shown	Missing view blocks architecture approval
Control matrix complete	Each material control has owner, activity, evidence, frequency and failure condition	Missing material control blocks release
Eval and test evidence ready	Thresholds, failures and defect disposition are documented	Unresolved high-risk failure blocks release
Evidence packet indexed	Evidence ids and versions are linked to claims and controls	Unindexed evidence returns to owner
Challenge memo complete	Second-line position, conditions and objections are documented	Missing memo limits gate to design discussion
Issue/action log current	Blockers, conditions, accepted issues and overdue actions are visible	Unknown blockers defer decision
Operational readiness confirmed	SOP, training, support, fallback and escalation are ready	Missing readiness reduces scope or blocks
ADR updated	Material tradeoffs, decision and review triggers recorded	Missing ADR blocks architecture sign-off
Material change rule defined	Model, prompt, RAG, tool, vendor and scope changes trigger review	Missing rule blocks scale
Post-release cadence set	Evidence, issue, quality and scale review cadence defined	Missing cadence turns go into no-go

No-go signals:

No named first-line owner.
Second-line objection unresolved and not escalated.
Critical or high blocker lacks closure plan.
Evidence packet cannot support the proposed gate decision.
Architecture lacks evidence capture for material control.
Conditional go has no owner, due date or closure evidence.
Internal audit is being asked to approve release.

12. Executive Narrative

Use this narrative for steering committee or portfolio review. Keep the wording factual and avoid saying "risk approved."

We are requesting conditional go for the AML copilot at the release gate.

The first-line accountable owner is the Financial Crime Operations Director, who owns the business outcome, operational process, control operation and residual-risk proposal. The use case is limited to L2 analyst support for alert types A and B. The AI is not permitted to close alerts, change disposition, file SARs or send customer communications.

The release packet includes the AML copilot architecture view, control matrix, eval/test evidence, second-line challenge memo, issue/action log and ADR-AML-COPILOT-004. Second-line functions reviewed the packet and recorded conditional no objection. The material conditions are paragraph-level source-span evidence for saved narratives and an ADR update for corpus, prompt, workflow and vendor material-change triggers, each with owner, due date and closure evidence.

The main value case is reduced narrative preparation time and improved case evidence consistency. The main risk/control considerations are unsupported narrative claims, analyst over-reliance and material corpus change. Open issues are classified as two release conditions and no unresolved blocker. The decision requested is conditional production release for the defined population, with review triggers for corpus update, prompt change, vendor gateway change, source-span sample failure, incident signal or scale request.

Internal audit is not approving this release. The use case will be visible to audit for future assurance planning because it affects financial crime case records.

Scenario-specific examples:

Use case	Executive narrative emphasis
GenAI contact center	Internal agent-assist scope, regulated content escalation, transcript evidence and quality sampling
AML copilot	Analyst-owned final disposition, source-span narrative evidence and typology coverage
Credit decision support	Decision-support boundary, policy consistency and adverse-action / fairness challenge route
KYC onboarding	Human review for rejection, missing-info workflow, false reject / accept issue route
AI vendor model	Data-use terms, model update notice, fallback and exit plan
Enterprise knowledge assistant	Approved-source governance, entitlement-aware retrieval and sensitive-corpus restrictions

13. Interview Drills

Drill 1: Explain Three Lines for AI in 30 seconds

Good answer:

For AI, Three Lines governance means first-line product and operations teams own the business outcome, controls and residual-risk proposal; second-line risk and compliance teams challenge policy fit, control design and evidence sufficiency; third-line internal audit provides independent assurance and does not approve release. The practical architecture artifact is a lifecycle gate model with decision records, evidence packets, challenge memos, issue/action logs and ADRs.

Drill 2: What is the most common failure?

Good answer:

The most common failure is saying "risk signed off" without defining who owns the release decision and residual risk. That blurs accountability. I would separate first-line decision, second-line challenge position and third-line assurance role in the gate record, then bind the decision to exact evidence and issue dispositions.

Drill 3: How would you apply this to an AML copilot?

Good answer:

The first line is financial crime operations, which owns analyst workflow, narrative quality and final disposition. Second line financial crime compliance challenges typology coverage, SAR boundary, source evidence and escalation rules. Internal audit does not approve release; it may later test whether governance and controls operated. The release packet should include source-span evidence for narratives, analyst approval events, issue/action log, eval coverage and an ADR stating the copilot cannot close alerts or file SARs.

Drill 4: How do you keep this from becoming bureaucracy?

Good answer:

I would make the model risk-tiered. Low-risk internal knowledge use gets a light gate. High-risk customer-impacting, credit, AML, KYC or agentic tool-use scenarios get full evidence forum and challenge memo. The key is reusable platform evidence: policy decision logs, trace events, control matrix templates and ADRs. That reduces meeting load while improving assurance readiness.

Drill 5: What would you ask the CTO to fund?

Good answer:

I would ask for an enterprise AI governance control plane: AI inventory, policy decision logging, evidence collector, trace store, issue/action register, ADR integration and material-change detection for model, prompt, RAG corpus, tools and vendor dependencies. This turns Three Lines from manual committee work into platform-supported governance.

14. Source Anchors

Source	Official link	Execution translation
IIA Three Lines Model	https://www.theiia.org/en/content/position-papers/2020/the-iias-three-lines-model-an-update-of-the-three-lines-of-defense/	Separate management ownership, risk/compliance challenge and internal audit assurance.
Basel Committee Corporate Governance Principles for Banks	https://www.bis.org/bcbs/publ/d328.htm	Align AI governance with banking governance, risk management, control environment and internal audit expectations.
COSO Internal Control Overview	https://www.coso.org/guidance-on-ic	Translate AI controls into control environment, risk assessment, control activities, information/communication and monitoring language.
NIST AI Risk Management Framework	https://www.nist.gov/itl/ai-risk-management-framework	Use Govern / Map / Measure / Manage to structure AI risk lifecycle and management action.
ISO/IEC 42001 AI Management System	https://www.iso.org/standard/81230.html	Treat AI governance as a management system with scope, operation, performance evaluation and improvement.
ISO/IEC 23894 AI Risk Management	https://www.iso.org/standard/77304.html	Anchor AI risk management, treatment and review vocabulary.