返回 Papers
AI 扩展计划 / Playbooks

AI Regulatory Horizon / Obligation Intelligence Playbook

AI Regulatory Horizon / Obligation Intelligence 是一套能力:

726AI_REGULATORY_HORIZON_OBLIGATION_INTELLIGENCE_PLAYBOOK.md

AI Regulatory Horizon / Obligation Intelligence Playbook

定位: 面向高级 AI PM / AI BA / CBAP+ / Product Governance / Enterprise Architect / Solution Architect / Compliance Technology / Model Risk / Financial Retail AI Owner 的主动监管地平线扫描与义务智能架构手册。 核心问题: AI laws、guidance、regulatory speeches、supervisory priorities、standards、enforcement patterns 和 vendor notices 持续变化。组织需要把这些外部变化转成 obligation-to-control architecture, 而不是等 exam、incident 或 audit finding 后临时响应。 重要说明: 本文是学习、作品集和企业治理设计材料,不构成法律意见、合规结论、模型验证意见、审计意见或监管解释。正式项目必须由 Legal、Compliance、Risk、Model Risk、Privacy、Security、Internal Audit、Business Owner、Technology Owner、Data Owner 和管理层确认。


1. Executive Framing

1.1 一句话定义

AI Regulatory Horizon / Obligation Intelligence 是一套能力:

External regulatory signals
-> applicability triage
-> obligation extraction
-> obligation ontology
-> impact mapping
-> owner assignment
-> control / eval / change linkage
-> evidence capture
-> management reporting

它回答八个问题:

  1. 哪些官方来源和监管信号正在变化。
  2. 哪些变化可能影响我们的 AI portfolio。
  3. 哪些 AI assets、产品、地区、客户旅程、供应商和控制受到影响。
  4. 哪些义务需要变成 control objective。
  5. 哪些义务需要变成 eval requirement。
  6. 哪些义务需要触发 release/change gate。
  7. 哪些 owner 必须在什么时限内做出 decision、remediation 或 risk acceptance。
  8. 管理层如何看到 exposure、coverage、overdue、evidence 和 residual risk。

1.2 与 Regulatory Response 的区别

Regulatory response 通常处理已经发生的事件:

exam request -> fact finding -> evidence pack -> response -> corrective action

Obligation intelligence 处理更早、更系统的问题:

new law / guidance / speech / standard / priority / incident
-> triage impact
-> update obligations
-> map controls and evals
-> trigger changes before harm or exam

高级表达:

We do not wait for a regulator, auditor or incident to tell us AI obligations changed. We run horizon scanning as an obligation intelligence architecture that turns external signals into internal controls, evals, release gates, evidence and management decisions.

1.3 金融零售为什么更难

LayerTypical obligation or expectation
AI-specific regulationrisk classification、provider/deployer duties、transparency、human oversight、post-market monitoring
Banking supervisionmodel risk、operational risk、third-party risk、governance、board oversight
Consumer protectionfair treatment、UDAAP、adverse action、complaints、credit reporting、dispute rights
Data and privacyPII、purpose limitation、retention、cross-border transfer、data subject rights
Security and resiliencecyber risk、prompt injection、tool misuse、availability、incident response
Standards and assuranceNIST AI RMF、ISO/IEC 42001、internal control library、audit evidence
AI obligations change by jurisdiction, product, use case, customer segment, AI role, automation level, data category, vendor dependency and release state. A single global answer is usually wrong.

2. Source Anchors

以下官方来源是本文的核心锚点。访问日期按 2026-06-30 记录。正式使用时必须复核最新版本、适用性和机构内部解释。

AnchorOfficial linkPlaybook usage
EU AI Act, Regulation (EU) 2024/1689https://eur-lex.europa.eu/eli/reg/2024/1689/oj/engAI obligations、risk-based classification、provider/deployer responsibilities、transparency、high-risk AI、post-market monitoring 和 incident reporting
NIST AI Risk Management Frameworkhttps://www.nist.gov/itl/ai-risk-management-framework用 Govern / Map / Measure / Manage 组织 AI risk management、control translation、measurement and management
ISO/IEC 42001https://www.iso.org/standard/42001用 AI Management System 组织 policy、leadership、planning、support、operation、performance evaluation、improvement
Federal Reserve SR 26-2https://www.federalreserve.gov/supervisionreg/srletters/SR2602.htmRevised model risk management guidance。SR 26-2 于 2026-04-17 supersedes SR 11-7 和 SR 21-8
CFPB circulars and guidance indexhttps://www.consumerfinance.gov/compliance/circulars/Consumer finance circulars、advisory opinions、bulletins、interpretive rules 和 policy statements 的监控入口

2.1 Source use discipline

  • 不把 source anchor 当作完整法律清单。
  • 不把法律文本直接复制成工程任务。
  • 不在产品文档中替 Legal / Compliance 下最终法律结论。
  • 每个来源必须记录 URL、source date、access date、authority type、jurisdiction、owner 和 version。
  • 每个 material signal 必须 resolve 成 no-impact rationale、watch item、obligation、control update、eval update、change request、risk acceptance 或 management escalation。

2.2 SR 26-2 nuance

SR 26-2 superseded SR 11-7 and SR 21-8, but GenAI and agentic AI should be treated through broader AI governance rather than blindly forced into legacy model-risk framing. Better stance:

  • Predictive models and statistical decision models still need model risk management.
  • GenAI, RAG and agentic AI combine model behavior with prompts, knowledge sources, tool permissions, workflow, human oversight, monitoring, vendor dependency and change risk.
  • Some components route to Model Risk; the whole AI system also routes to product governance, data governance, third-party risk, security, EvalOps, release governance and evidence architecture.
  • Obligation intelligence should classify the obligation and route it to the right governance domain.

3. Source Taxonomy

3.1 Source categories

Source categoryExampleWhy it mattersDefault owner
Binding law / regulationEU AI Act, national AI lawcreates explicit legal obligations and deadlinesLegal / Compliance
Supervisory guidanceSR letters, interagency guidanceshapes examiner expectations and risk managementCompliance / Risk
Regulatory circular / bulletinCFPB circulars, policy statementsinterprets consumer finance expectationsCompliance / Product Risk
Regulatory speech / testimonypublic remarks, hearingsearly signal of supervisory directionCompliance Intelligence
Supervisory prioritiesexam priorities, risk alertsshows what reviews may emphasizeRisk / Internal Audit
StandardsNIST AI RMF, ISO/IEC 42001converts principles into management system practiceArchitecture / AI Governance
Enforcement / litigationconsent orders, court decisionsreveals concrete failure modesLegal / Compliance
Industry incidentspeer AI failure, vendor outagetriggers scenario evals and control challengeOperational Risk / PM
Vendor noticesmodel, SLA, data terms, regioncan change behavior or data boundaryVendor Owner / TPRM
Internal findingsaudit issue, validation issue, complaint trendconverts internal evidence into control updatesControl Owner

3.2 Source authority levels

LevelMeaningAction
Level 1: Bindinglaw, regulation, court order, formal rulemandatory triage and action plan if applicable
Level 2: Supervisoryguidance, circular, exam priority, SR letterimpact assessment for relevant business lines
Level 3: StandardNIST, ISO, industry control standardmap to management system and control library
Level 4: Signalspeech, consultation, peer incident, vendor notewatch item or scenario trigger
Level 5: Contextcommentary, conference summaryresearch input unless confirmed by stronger source

3.3 Source registry fields

FieldDescriptionExample
source_idstable source identifierSRC-EUAI-2024-1689
titleofficial title or internal nameRegulation (EU) 2024/1689
authority_typelaw, guidance, circular, standard, speech, incident, vendor noticelaw
issuing_bodyregulator, standards body, vendor, internal auditEuropean Union
jurisdictionEU, US federal, state, global, internalEU
urlofficial linkhttps://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng
source_datepublication, effective or notice date2024-07-12 Official Journal publication
access_datedate reviewed by organization2026-06-30
statusactive, superseded, consultation, watch, archivedactive
source_ownerresponsible monitoring roleAI Regulatory Intelligence Lead
materialitylow, medium, high, criticalhigh

3.4 Signal types

Signal typeMeaningExample action
New obligationcreates or clarifies required actioncreate obligation object and control gap assessment
Timing changechanges deadline or phase-in dateupdate calendar and roadmap
Scope changechanges covered entities, products, roles, data or use casesre-run applicability triage
Supervisory expectationindicates examiner expectation without direct rule textadd control challenge or briefing
Control practicesuggests control, documentation or testing approachupdate control library
Risk themehighlights emerging harm or failure modeadd eval scenarios and KRI
Evidence expectationindicates records, logs, reports or documentation neededupdate evidence binder design
No-impact updatereviewed and not applicablerecord rationale and next review date

4. Obligation Ontology

4.1 Why ontology matters

Without an ontology, regulatory work becomes unstructured notes:

legal memo -> spreadsheet -> meeting -> partial control update -> lost context

With an ontology, the organization can query:

  • Which Tier 1 AI assets are impacted by EU AI Act high-risk obligations.
  • Which controls support transparency obligations but lack evidence.
  • Which SR 26-2 signals touch model validation versus broader GenAI governance.
  • Which CFPB circulars affect credit, complaint, dispute or adverse action journeys.
  • Which obligation objects are overdue, ownerless or waiting for Legal interpretation.

4.2 Obligation object model

FieldDescription
obligation_idstable ID such as OBL-EUAI-ART50-DISCLOSURE-001
source_idsource registry ID
source_quote_refarticle, section, paragraph, page or memo anchor
jurisdictionapplicable jurisdiction
authority_levelbinding, supervisory, standard, signal, context
obligation_typeinventory, risk classification, data governance, transparency, logging, human oversight, eval, incident, vendor, reporting, training
action_verbidentify, document, assess, test, monitor, disclose, report, retain, train, escalate
applicability_conditionwhen the obligation may apply
organization_roleprovider, deployer, creditor, vendor user, model developer, system operator
impacted_personcustomer, applicant, employee, merchant, analyst, investigator, regulator
impacted_artifactAI asset, model, prompt, RAG corpus, tool, process, UI, policy, evidence
risk_tier_hintTier 0, Tier 1, Tier 2, Tier 3, legal review required
control_objective_idlinked internal control objective
control_activity_idlinked control implementation
eval_requirement_idlinked eval or test
change_trigger_idlinked release/change trigger
evidence_requirement_idlinked evidence artifact
owner_roleaccountable role for implementation
legal_statuspending interpretation, confirmed applicable, confirmed not applicable, conditional
implementation_statusnew, mapped, in progress, implemented, monitored, retired
review_cadenceevent-driven, monthly, quarterly, annual

4.3 Obligation type taxonomy

Obligation typeProduct / architecture translation
AI inventoryregister use case, system, model, vendor, data, automation level, risk tier, owner and status
Risk classificationroute use case to prohibited, high-risk, high-impact, controlled, low-risk or no-go path
Data governancedata lineage, source authority, quality, minimization, purpose limitation, retention
Technical documentationarchitecture package, system card, model card, eval report, limitations, intended use
TransparencyUI disclosure, user notice, AI interaction clarity, generated content labeling
Human oversightreviewer authority, override, escalation, training, sampled QA, workload capacity
Accuracy / robustnessevaluation, monitoring, regression tests, uncertainty and failure mode management
Cybersecuritythreat model, prompt injection, tool authorization, data exfiltration prevention
Logging and traceabilityversioned logs, trace IDs, prompt/model/index/tool versions, retention
Post-market monitoringdrift, complaint, incident, override, vendor change, control effectiveness
Serious incidentseverity, containment, reporting decision, customer remediation, regulator path
Third-party AIvendor due diligence, contract clauses, model update notice, exit plan
AI literacy / trainingrole-based competency, training evidence, attestations, refresh cadence
Management reviewKPI/KRI reporting, residual risk, exceptions, corrective actions

4.4 Relationship rules

Source -> Signal -> Obligation -> Applicability decision -> AI asset
-> Control objective -> Control activity -> Eval requirement
-> Change request -> Evidence artifact -> Management report
  • No obligation without a source.
  • No applicable obligation without an owner.
  • No control objective without test or evidence expectation.
  • No material obligation update without change impact screening.
  • No closed obligation without implementation status and evidence link.

5. Applicability Triage

5.1 Triage principles

Applicability triage is not a legal conclusion. It is a structured fact-finding and routing process that lets Legal/Compliance make the final interpretation efficiently. Good triage separates external source language, internal interpretation hypothesis, business facts, technical facts, open legal questions, control implications and evidence implications.

5.2 Applicability dimensions

DimensionQuestions
JurisdictionWhere are customers, users, employees, entity, data, model provider and service delivered?
ProductCredit, deposit, payment, wealth, insurance, AML, KYC, customer service, marketing, HR?
Use caseWhat business task does AI support, draft, recommend, decide or act on?
Organization roleProvider, deployer, creditor, model developer, vendor user, processor or operator?
AI capabilityPredictive scoring, classification, ranking, RAG, generation, agent, optimization, tool use?
Automation levelRead, summarize, draft, recommend, decide, execute, write, notify?
Customer impactAccess, price, service, complaint, dispute, account action, investigation or communication?
Data categoryPublic, internal, PII, financial, credit, AML, special category, children, confidential?
Vendor dependencyExternal LLM, cloud, model provider, vector DB, case platform, analytics vendor?
Evidence needDoes a regulator, auditor, customer or employee need explanation or traceability?

5.3 Triage output categories

CategoryMeaningRequired action
Applicablesource likely applies to one or more AI assetscreate obligation objects, map controls, assign owners
Conditionally applicabledepends on role, jurisdiction, data, product or designrecord conditions and legal questions
Not applicable with rationalereviewed and does not apply to current scoperecord rationale, review trigger and approver
Watch itemnot binding or not clear, but potentially materialset review date and scenario trigger
Control enhancementno direct legal duty, but useful risk reductionadd to control backlog or standard
Management issuerequires funding, risk appetite or enterprise decisionescalate to governance forum

5.4 Triage workflow

Signal received
-> source owner confirms authenticity and version
-> compliance analyst classifies signal type
-> BA gathers product/process facts
-> architect gathers system/data/vendor facts
-> Legal/Compliance confirms applicability status
-> PM and control owner map impact and action
-> governance forum reviews high-impact items
-> obligation graph updated

5.5 Triage worksheet example

FieldFilled example
Signal IDSIG-CFPB-2026-CIRCULAR-REVIEW-001
SourceCFPB circulars and guidance index
Business areaCredit card dispute and fee servicing
AI assetAI-CS-RAG-042 customer service assistant
JurisdictionUS retail banking, EU channel screened separately
AI behaviorRAG answer + LLM draft + escalation recommendation
Customer impactMedium to high because fee, dispute, complaint and account guidance can affect customer rights
Data usedAccount metadata, customer message, approved policy knowledge base
Initial applicabilityConditionally applicable, depending on specific circular topic
Legal questionsDoes the circular affect dispute rights, adverse action, credit reporting or complaint routing?
Control implicationAdd circular-specific scenario to eval set and escalation rules if applicable
OwnerCompliance Product Risk Lead with Customer Service PM

6. Horizon Scanning Cadence

6.1 Cadence by source

CadenceSourcesOutput
Dailyurgent regulator updates, vendor model notices, severe incidentsimmediate high-materiality signal ticket
Weeklyofficial regulatory pages, CFPB circulars index, SR letters, standards announcementsweekly horizon log and triage queue
Monthlyspeeches, supervisory priorities, enforcement summaries, peer incidentsmonthly regulatory horizon brief
QuarterlyNIST/ISO framework changes, control library refresh, AIMS reviewquarterly obligation dashboard and management review
Event-drivennew regulation, exam letter, major vendor change, serious AI incidentwar-room style impact assessment

6.2 Weekly scan agenda

StepActivityOutput
1validate official source updatessource registry update
2classify each signalsignal type and authority level
3identify likely impacted domainsproduct and use-case tags
4assign triage ownerowner and due date
5close no-impact items with rationaleno-impact decision record
6escalate high-impact signalsgovernance agenda item

6.3 Monthly review decisions

Agenda itemDecision
New high-materiality sourcesaccept for action, watch, or close
Open applicability questionsescalate, request legal opinion, or time-box research
Overdue obligation ownersextend with rationale or escalate
Tier 1 / Tier 2 exposureupdate control and eval backlog
Control library changesapprove new or revised controls
Eval library changesadd, retire or recalibrate eval scenarios
Release governance triggerscreate change tickets or hold releases
Management reportingapprove dashboard narrative and risk decisions

6.4 Quarterly and event-driven triggers

7. Obligation-to-Control/Eval/Change Graph

7.1 Why a graph

Spreadsheets help intake, but graph thinking answers multi-hop questions:

  • Which obligations affect all credit AI use cases in the EU.
  • Which controls support both EU AI Act logging and internal model risk evidence.
  • Which eval suites must run after a vendor model change.
  • Which open obligations block a release.
  • Which obligations have evidence but no production monitoring.

7.2 Node types

NodeExample
SourceEU AI Act, SR 26-2, CFPB circular index, NIST AI RMF, ISO/IEC 42001
Signalnew obligation, revised guidance, standard update, speech, vendor notice
Obligationtransparency disclosure, logging, human oversight, data governance
AI assetAI-CREDIT-EXPLAIN-001, AI-AML-NARRATIVE-002
Business capabilityconsumer credit, AML investigation, customer servicing
Productcredit card, deposit account, payment dispute
Control objectiveensure human oversight is effective
Control activityunderwriter must approve and record override reason
Eval requirementreason-code consistency eval, groundedness eval, prompt injection test
Change requestprompt update, RAG corpus update, tool permission change
Evidenceeval report, trace sample, approval record, dashboard snapshot
Ownerproduct owner, compliance owner, architect, control owner

7.3 Edge types

EdgeMeaning
source_generates_signalsource update created a signal
signal_extracts_obligationsignal produced an obligation or expectation
obligation_applies_to_assetobligation potentially or confirmed applies to AI asset
asset_supports_capabilityAI asset belongs to a business capability
obligation_requires_controlobligation maps to control objective
control_tested_by_evalcontrol effectiveness depends on an eval
control_evidenced_by_artifactevidence proves design or operation
obligation_triggers_changeobligation update requires system/process release
change_requires_evalrelease gate must include eval
owner_accountable_for_nodeowner has accountable responsibility

7.4 Minimum graph queries

QueryManagement use
Show all AI assets impacted by EU AI Act high-risk obligationsportfolio exposure
Show all obligations with no owneraccountability gap
Show Tier 1 assets with obligations but no eval linkrelease gate gap
Show controls linked to SR 26-2 and GenAI assetsmodel risk and AI governance boundary review
Show CFPB signals linked to customer-facing AIconsumer harm prevention
Show obligations overdue by more than 30 daysmanagement escalation
Show releases blocked by unresolved obligationsrelease governance
Show evidence coverage by control objectiveaudit readiness

7.5 Implementation pattern

Start simple:

Markdown templates + stable IDs + spreadsheet relationships + control catalog links

Then mature:

GRC system + AI inventory + model registry + data catalog + release system + eval registry + evidence store

Use predictable IDs:

ObjectExample
SourceSRC-EUAI-2024-1689
SignalSIG-EUAI-HIGHRISK-2026-001
ObligationOBL-EUAI-HR-LOGGING-001
Control objectiveCTRL-OBJ-AI-LOG-001
EvalEVAL-GROUNDING-CREDIT-001
Change requestAI-CR-2026-0715-RAG-CORPUS
EvidenceEV-AI-CREDIT-EXPLAIN-REL-014

8. Governance Workflow

8.1 Workflow stages

StagePurposeExit criteria
Source intakeconfirm official source, version and authoritysource record created
Signal triagedecide materiality and likely business impactsignal classified and owner assigned
Applicability triagedetermine whether source may applyapplicability decision or legal question
Obligation extractionconvert source into structured obligation objectsobligation records created
Impact mappinglink obligations to assets, controls, evals, changes and evidencegraph edges complete
Owner assignmentassign accountable and consulted rolesRACI and due date recorded
Control updateupdate control library and operating procedureapproved control change
Eval updateupdate eval scenario, threshold or regression gateeval registered and scheduled
Release linkagecreate or update change request if behavior must changechange ticket linked
Evidence capturedefine and collect evidenceevidence link and retention rule complete
Management reviewreport status, risk and decisionsdashboard reviewed and decisions logged

8.2 Decision states

StateMeaning
Newsource or signal recorded, not yet triaged
Triage in progressowner is gathering facts
Legal interpretation pendingapplicability depends on formal interpretation
Mapped to controlsobligation has control objective and control activity
Implementation in progresscontrol, eval, process or system update underway
Implementedrequired action completed and evidence linked
Monitoredproduction monitoring or periodic review active
Accepted riskmanagement accepted residual risk with rationale
Closed no impactclosed with documented rationale and review trigger
Retiredobligation superseded or no longer applicable

8.3 Forums and escalation

ForumScopeCadence
Regulatory Horizon Triagenew signals, materiality, owner routingweekly
AI Product Risk Reviewuse-case impact, customer harm, roadmap and releaseweekly or biweekly
Architecture and Control Reviewcontrol library, eval links, evidence architecturebiweekly
Material AI Governance CouncilTier 1 obligations, gaps, risk acceptance, fundingmonthly or event-driven
AIMS Management ReviewISO/IEC 42001-aligned effectivenessquarterly
Escalate when a signal may affect Tier 1 assets, block a release, create consumer harm, require new funding/tooling, change vendor terms, miss owner SLA, or require management risk acceptance.

8.4 Quality gate

CheckPass standard
Source traceabilityofficial source, access date and section reference recorded
Applicability clarityapplicable, conditional, not applicable or watch state documented
Control linkagecontrol objective and activity linked or no-control rationale approved
Evidence linkageevidence artifact, dashboard or retention plan linked
Owner accountabilityaccountable owner and review cadence recorded

9. Dashboards and KRIs

9.1 Executive dashboard

MetricDefinitionWhy it matters
Open material signalshigh or critical signals not dispositionedshows horizon load
Open applicability decisionsitems waiting for interpretation or factsshows bottleneck
Obligations by authority levelbinding, supervisory, standard, signalshows backlog risk profile
Tier 1 assets impactedhigh-impact assets linked to open obligationsshows customer/regulatory exposure
Overdue obligation actionsowner actions past dueshows execution risk
Evidence coverageobligations with linked evidence over applicable obligationsshows audit readiness
Eval linkage coverageapplicable obligations with eval or test linksshows release gate maturity
Control gap countapplicable obligations without implemented controlsshows remediation need
Change linkage countobligations that triggered change requestsshows operational follow-through
Accepted residual riskobligations closed through risk acceptanceshows management decisions

9.2 Product and control dashboards

View / KRITrigger or use
Obligations by product lineidentify products with largest regulatory horizon load
Obligations by journeylocate onboarding, servicing, complaint, credit or payment hot spots
Obligations by releaseblock or condition releases with unresolved obligations
Obligations by AI capabilitysee RAG, agent, scoring, generation or tool-use concentration
Obligations by jurisdictionprevent one-size-fits-all rollout
Applicable obligation without controlany Tier 1 obligation lacks control within 10 business days
Control without evidencecontrol marked implemented but no evidence link
Eval gapobligation requires test but no eval exists
Change gapobligation implies system/process change but no change ticket exists
Evidence staleevidence older than review cadence or prior to material release

9.3 Board narrative

Good management reporting answers what changed externally, which changes matter, which assets are exposed, which obligations need decisions, which gaps are overdue, which residual risks were accepted, and which investments are required. Bad reporting only shows article counts, meeting counts, monitored-regulation counts, or green status without traceable evidence.

10. RACI

10.1 Operating RACI

ActivityLegalComplianceAI PMBAArchitectModel RiskPrivacy/SecurityDataVendorGovernance
Source authenticityCA/RCCCCCCCR
Signal classificationCA/RCCCCCCCR
Applicability interpretationA/RA/RCCCCCCCC
Product impactCCA/RRCCCCCC
Process impactCCCA/RCCCCCC
System/data/vendor factsCCCCA/RCCRRC
Obligation extractionCA/RCRCCCCCR
Control mappingCA/RCRA/RCCCCC
Eval linkageCCCCCA/RCCCC
Change ticket creationCCA/RRA/RCCCCC
Evidence designCCCRA/RCCCCA/R
Dashboard reportingCCCCCCCCCA/R
Risk acceptanceCCCCCCCCCA with executive owner
Legend: A = accountable, R = responsible, C = consulted.

10.2 Role expectations

RoleMust be able to explain
AI PMhow horizon changes affect roadmap, scope, release gates, customer value and risk
BA / CBAP+how obligations become requirements, process rules, exception paths, evidence and decision rights
Architecthow obligations map to boundaries, data flows, logs, evals, controls, change and evidence
Compliance / Legalwhich sources matter, what interpretation is confirmed, what remains conditional
Model Riskwhich components require model risk governance and how that interacts with broader AI governance
Governance Leadbacklog health, owner accountability, dashboard narrative and management decisions

11. Financial Retail Examples

11.1 Credit explanation assistant

Use case: AI assists underwriters and agents by summarizing credit policy, explaining factors and drafting customer communication. Final decision and formal adverse action reason remain controlled by approved workflow and human accountability.

Horizon signalImpact
EU AI Act high-risk and transparency screeningassess role, risk tier, documentation, logging and human oversight for EU scope
CFPB circulars affecting adverse action or credit reportingupdate reason-code eval and notice quality controls
SR 26-2route predictive scoring to model risk and RAG/LLM workflow to broader AI governance
NIST AI RMFconnect Govern/Map/Measure/Manage evidence to release and monitoring
ISO/IEC 42001include in AIMS management review and improvement cycle
Obligation areaControl
------
Explanation accuracyreason-code consistency eval and policy-grounded review
Human oversightunderwriter approves final reason and can override AI draft
Loggingtrace model version, prompt, policy source, draft and final reason
Changeprompt, model, policy source or notice template update triggers release gate
Consumer harmcomplaint and appeal themes monitored by segment

11.2 AML investigation narrative copilot

Use case: AI summarizes transactions, KYC profile, watchlist hits and case notes, then drafts investigation narrative without deciding SAR filing or case closure.

SignalControl response
supervisory focus on AML model and case qualityadd case quality and evidence-citation eval
SR 26-2classify transaction monitoring model separately from LLM narrative support
vendor model updatetrigger regression on narrative completeness, citation support and sensitive data handling
peer incidentadd red-team scenario for missing adverse evidence
Core controls: AI cannot close case; narrative cites case evidence IDs; investigator confirms final narrative; SAR decision boundary is explicit; prompt and model version are logged; case sampling tests completeness and overstatement.

11.3 Customer service RAG assistant

Use case: AI answers customer questions about fees, disputes, account servicing and product policy. High-risk paths route to human agent or supervisor.

SignalControl response
CFPB circulars index updatescreen for dispute, fee, complaint, credit reporting or deceptive practice implications
EU AI Act transparencyreview AI interaction disclosure for EU users
ISO/IEC 42001 management reviewinclude chatbot risk trend and corrective actions
vendor model safety policy changeupdate refusal, escalation and disclosure tests
Core controls: approved knowledge source allowlist; citation and effective-date display; complaint detection; prohibited statements list; approved disclosure; monitoring for complaint spike and deflection side effects.

11.4 Payment exception agent

Use case: AI triages failed payments, suggests next action and drafts internal notes. Write-enabled actions require human confirmation and policy-issued authorization token.

SignalControl response
operational resilience guidanceadd fallback, manual queue and recovery evidence
security guidancestrengthen tool authority, idempotency and rate limiting
consumer protection signalmonitor wrongful payment block or delayed dispute resolution
vendor outagereview concentration risk and exit plan
Core controls: separate read/draft/write/submit/approve permissions; idempotency key for writes; maker-checker for high-value or customer-adverse actions; policy decision ID in logs; tested rollback and compensation plan.

11.5 Wealth advisor copilot

Use case: AI summarizes client profile, approved product materials and meeting notes for advisor preparation. It does not send personalized recommendation directly to client.

SignalControl response
suitability / best interest supervisory themeupdate advisor approval and suitability checklist
AI transparency guidancereview client communication boundary
vendor data-use noticecheck client data processing, retention and training exclusions
standards updateupdate recordkeeping and evidence mapping
Core controls: approved product library only; suitability checklist remains human-owned; client communication archived; generated drafts marked as drafts; advisor edit and approval recorded.

12. Templates

12.1 Source Registry Template

FieldExample-filled entry
source_idSRC-FED-SR26-2
titleFederal Reserve SR 26-2: Model Risk Management Guidance
authority_typesupervisory guidance
issuing_bodyBoard of Governors of the Federal Reserve System
jurisdictionUS federal banking
official_urlhttps://www.federalreserve.gov/supervisionreg/srletters/SR2602.htm
source_date2026-04-17
access_date2026-06-30
statusactive, supersedes SR 11-7 and SR 21-8
materialityhigh for supervised banking organizations using material models
ownerModel Risk Policy Owner with AI Governance Lead
review_cadencequarterly and event-driven

12.2 Signal Intake Template

FieldExample-filled entry
signal_idSIG-FED-SR26-2-GENAI-ALIGNMENT-001
source_idSRC-FED-SR26-2
signal_typesupervisory expectation and governance alignment
summaryrevised model risk management guidance requires alignment with current model use and risk profile
likely_impacted_domainscredit risk, fraud, AML, pricing, forecasting, GenAI systems with model components
initial_materialityhigh
triage_ownerAI Governance Lead
triage_due_date2026-07-15
recommended_first_actionidentify AI assets with predictive model components and separate broader GenAI/agentic controls
decision_statustriage in progress

12.3 Applicability Triage Template

FieldExample-filled entry
AI assetAI-CREDIT-EXPLAIN-001
ProductCredit card underwriting support and explanation
JurisdictionUS and EU customer segments separated
Organization rolecreditor, deployer, vendor user
AI capabilitypredictive score, RAG policy assistant, LLM rationale draft
Automation levelrecommend and draft, no autonomous final decision
Customer impacthigh because credit access and explanation are affected
Data categoryPII, financial transaction data, credit attributes
Source being triagedEU AI Act and SR 26-2
Applicability resultconditional pending Legal confirmation for EU high-risk path; model risk applies to predictive component
Control implicationhigh-risk screen, reason-code eval, trace logging, human oversight, model validation, broader AI governance controls
Decision ownerCredit Risk Executive with Legal, Compliance and AI Governance

12.4 Obligation Object Template

FieldExample-filled entry
obligation_idOBL-EUAI-HR-HUMAN-OVERSIGHT-001
source_idSRC-EUAI-2024-1689
authority_levelbinding law
obligation_typehuman oversight
action_verbdesign, implement, document and monitor
applicability_conditionAI system is assessed as high-risk or high-impact decision support for EU credit use case
impacted_artifactunderwriter workflow, UI, training, logs, QA process
control_objective_idCTRL-OBJ-HO-001
eval_requirement_idEVAL-HO-EFFECTIVENESS-001
evidence_requirement_idEV-HO-REVIEW-LOG-001
owner_roleCredit Operations Owner
legal_statusconditional pending Legal confirmation
implementation_statusmapped to controls
review_cadencemonthly operational sample and quarterly governance review

12.5 Obligation-Control-Eval-Change Map

ObligationControl objectiveControl activityEval / testChange triggerEvidence
Customer-facing AI disclosureuser understands AI interaction and escalation pathapproved UI copy and channel-specific disclosureUX comprehension review and compliance approvalnew channel, new geography, major UX redesignscreenshot, copy approval, release record
Logging and traceabilityAI output can be reconstructed and auditedlog model, prompt, index, policy source, tool call, human reviewtrace completeness samplemodel/prompt/index/tool changetrace sample, retention proof
Human oversighthuman can intervene with context and authorityreviewer sees evidence, edits, overrides and escalatesoversight effectiveness testautomation level increases or queue changesreview log, override analytics
Consumer harm preventionAI does not mislead customers on fees, disputes or creditprohibited claim rules and escalation pathsscenario eval and complaint trend reviewCFPB circular or policy updateeval report, complaint dashboard
Vendor model changethird-party AI change does not break controlsvendor notice review and regression gatedomain regression and security reviewvendor model, region, data term or SLA changevendor notice, eval run, approval

12.6 Management Report Template

# AI Regulatory Horizon Management Update
Reporting period: 2026 Q3 first month
Prepared for: AI Governance Council and Risk Committee
External changes reviewed:
- 2 binding-law sources reviewed.
- 3 supervisory guidance items reviewed.
- 4 consumer finance circular or bulletin items screened.
- 5 standard or framework changes monitored.
- 2 vendor notices assessed.
Material decisions:
- EU AI Act high-risk screening remains mandatory for EU credit and employment-related AI assets.
- SR 26-2 alignment review opened for predictive model components inside GenAI-enabled workflows.
- CFPB circular monitoring added to customer service, dispute, credit reporting and fee journeys.
Portfolio impact:
- 7 Tier 1 AI assets linked to open obligations.
- 3 releases require updated eval gates before production expansion.
- 2 vendor contracts require AI update-notice review.
Top risks:
- Open Legal interpretation on EU deployer role for one cross-border customer service assistant.
- Evidence coverage below target for human oversight logs in one credit workflow.
- Vendor model update cadence not contractually aligned with release governance.
Decisions requested:
- Approve funding for obligation graph integration with AI inventory and release system.
- Assign executive owner for cross-jurisdiction AI applicability policy.
- Accept temporary residual risk for one Tier 2 chatbot pilot with limited traffic and enhanced monitoring.

12.7 Evidence Binder Template

FolderRequired evidence
01-sourceofficial source PDF/link, access record, source registry row
02-signalsignal intake, classification, materiality rationale
03-applicabilitytriage worksheet, legal questions, interpretation decision
04-obligationsobligation objects, ontology tags, status
05-impactaffected AI assets, products, controls, evals, vendors
06-controlscontrol objectives, control activities, owners
07-evalseval scenarios, thresholds, results, failure analysis
08-changechange tickets, release gates, approvals, rollback plan
09-evidencelogs, dashboards, screenshots, approvals, test artifacts
10-reportingmanagement memo, risk acceptance, committee minutes

13. 30-Day Lab

Week 1: Source and signal foundation

DayOutputAcceptance standard
Day 1Source registryfive official anchors recorded with URL, owner, access date
Day 2Source taxonomyauthority type, jurisdiction and review cadence defined
Day 3Signal intake templatecaptures type, materiality, affected domain, owner and due date
Day 4Ten sample signalsincludes law, guidance, standard, circular, vendor and incident categories
Day 5Weekly triage agendaincludes classify, assign, close, escalate and dashboard update

Week 2: Ontology and triage

DayOutputAcceptance standard
Day 6Obligation object modelincludes source, applicability, action verb, owner, control, eval, change and evidence
Day 7Applicability dimensionsjurisdiction, product, use case, role, AI capability, automation, data, vendor and customer impact
Day 8Triage worksheetone example for customer service RAG and one for credit explanation
Day 9Decision categoriesapplicable, conditional, no-impact, watch, control enhancement and management issue
Day 10Legal questions packprecise Legal/Compliance questions, not vague requests

Week 3: Graph and controls

DayOutputAcceptance standard
Day 11Obligation-to-control mapat least 12 obligation objects linked to controls
Day 12Eval linkage mapeach high-impact obligation has an eval or test
Day 13Change linkage mapmaterial obligations trigger release/change tickets
Day 14Evidence binder mapeach control has named evidence and retention rule
Day 15Graph query listat least eight management queries documented

Week 4: Reporting and operating model

DayOutputAcceptance standard
Day 16RACIroles for Legal, Compliance, PM, BA, Architect, Model Risk, Privacy/Security, Data, Vendor and Governance
Day 17Dashboard specexecutive, product and control dashboards with KRIs
Day 18Governance workflowstages from source intake to management review
Day 19Financial retail case packcredit, AML, customer service, payment and wealth examples
Day 20Management memoone-page report with external changes, impact, top risks and decisions

Days 21-30: Portfolio simulation

DayOutputAcceptance standard
Day 21Scenario chosencredit explanation, AML narrative, customer service RAG, payment agent or wealth copilot
Day 22Source registry populatedall anchors have official links and access dates
Day 2312 obligation objectseach has owner, control, eval, change and evidence links
Day 24Impact graphsource, obligation, asset, control, eval, change and evidence connected
Day 25Mock triage forumdecisions recorded for applicable, conditional, watch and no-impact items
Day 26Release gate updatedat least three obligations alter eval or release criteria
Day 27Dashboard builtoverdue, coverage, Tier 1 impact, evidence and residual risk views
Day 28Board memo writtentop three horizon signals and decisions requested
Day 29Self-review completedno orphan obligation, no ownerless item, no evidence-free closure
Day 30Interview rehearsal30-second and 2-minute explanations with one financial retail example

14. Interview Answers

14.1 30-second answer

AI regulatory horizon scanning should not be a legal newsletter. I would build an obligation intelligence architecture. It monitors official laws, guidance, circulars, speeches, supervisory priorities, standards and vendor notices; triages applicability by jurisdiction, product, use case, role, AI capability, automation and customer impact; extracts structured obligation objects; and maps each obligation to controls, evals, change tickets, evidence and owners. That gives management a real view of regulatory exposure before an exam, incident or release failure.

14.2 2-minute answer

In a regulated financial institution, AI obligations change by jurisdiction, product, use case, customer segment, automation level, data and vendor dependency. My approach starts with a source taxonomy and source registry. EU AI Act, NIST AI RMF, ISO/IEC 42001, SR letters and CFPB circulars are official source anchors with version, access date, authority level and owner. When a signal appears, I do applicability triage. The BA gathers process and customer impact facts, the architect gathers system, data, model, RAG, tool and vendor facts, and Legal/Compliance confirms interpretation. If material, the signal becomes an obligation object with action verb, applicability condition, owner, control objective, eval requirement, change trigger and evidence requirement. The important architecture is the graph: source to signal to obligation to AI asset to control to eval to change to evidence. That lets us answer which Tier 1 assets are impacted, which releases are blocked, which controls lack evidence and which obligations are overdue. For SR 26-2, I would not blindly treat every GenAI or agentic system as a legacy model risk item. Predictive model components still route to model risk, but RAG, prompts, tool authority, workflow, human oversight, vendor change, EvalOps, incident and evidence controls require broader AI governance.

14.3 English version

I would run AI regulatory horizon scanning as an obligation intelligence system, not as a passive monitoring newsletter. The system captures official sources, classifies signals, triages applicability by jurisdiction, product, use case, role, AI capability, automation level, data and customer impact, then extracts structured obligation objects. Each obligation is mapped to an AI asset, control objective, control activity, eval requirement, change trigger, evidence artifact and accountable owner. The architecture is a graph, because management needs to know which high-impact AI assets are exposed, which releases are blocked, which controls lack evidence and which residual risks require acceptance.

14.4 Deep-dive Q&A

QuestionStrong answer
How is this different from regulatory response?Regulatory response is event-driven, triggered by exam, incident or audit. Obligation intelligence turns external changes into controls, evals, releases and evidence before a failure.
How do you avoid over-governance?Use authority level, materiality, risk tier and applicability conditions. Low-risk signals can become watch or no-impact records; Tier 1 assets get enhanced controls.
How do you handle conflicting jurisdictions?Segment by jurisdiction, entity, product, customer location, data location and AI role. The graph can represent different applicability states for the same asset.
What makes a good obligation object?It has source, action verb, applicability condition, owner, control objective, eval link, change trigger, evidence requirement, status and review cadence.
How does NIST AI RMF fit?NIST provides Govern/Map/Measure/Manage language for risk management, but does not replace legal applicability analysis.
How does ISO/IEC 42001 fit?ISO/IEC 42001 gives the AI management system structure. Obligation intelligence feeds that management system.
What is the SR 26-2 nuance?It supersedes SR 11-7 and SR 21-8, so model risk programs need updating. GenAI and agentic systems also need broader AI governance.
What artifact would you show?A source registry, applicability matrix, obligation ontology, graph diagram, control/eval/change map, dashboard, RACI and management memo.

15. Pitfalls and Self-Assessment

PitfallBetter practice
Treating horizon scanning as news monitoringrequire disposition for every material signal
One global applicability answersegment applicability by jurisdiction, product and use case
Copying legal text into PRDextract action verb, condition, control and evidence
No distinction between obligation and controldefine control activity and evidence
No eval linkagemap obligations to eval and regression suites
No change linkagetrigger release/change ticket when behavior must change
No ownerassign accountable role and SLA
No evidence retentiondesign evidence binder and retention rule
Overusing model risk for GenAIroute components to model risk and broader AI governance
Dashboard vanity metricsreport Tier 1 impact, overdue actions, control gaps and evidence coverage
Self-assessment checklist:
  • Source registry includes official URLs, access dates, authority levels and owners.
  • Source taxonomy covers laws, guidance, circulars, speeches, priorities, standards, incidents, vendor notices and internal findings.
  • Applicability triage uses jurisdiction, product, use case, role, AI capability, automation, data, vendor and customer impact.
  • Obligation ontology includes source, action verb, applicability condition, owner, control, eval, change and evidence links.
  • Obligation graph can answer exposure, owner, control, eval, evidence and release-blocking queries.
  • Horizon cadence includes weekly scan, monthly review, quarterly management review and event-driven triggers.
  • Dashboards show materiality, Tier 1 impact, overdue actions, evidence coverage and residual risk.
  • RACI separates Legal interpretation, product impact, process analysis, architecture mapping, model risk and governance reporting.
  • Financial retail cases cover credit, AML, customer service, payment and wealth.
  • SR 26-2 is handled with both model-risk alignment and broader AI governance boundaries.

16. Portfolio Artifact Pack

ArtifactStandard
Source registryfive official anchors with version, access date and owner
Source taxonomyauthority levels and source categories
Signal logat least ten realistic signals with materiality and owner
Applicability triagetwo jurisdictions and two products analyzed
Obligation ontologytwelve obligations with complete fields
Graph diagramsource-to-evidence relationships visible
Control/eval/change mapevery high-impact obligation has a control, eval and change trigger
Dashboardexecutive, product and control views
RACIaccountable roles and escalation path
Management memotop signals, portfolio impact, decisions and residual risk
The portfolio should prove one skill:
You can convert changing AI regulation and supervisory expectations into operating architecture.