Implementation Guide | Banking
AI exam readiness timeline for banks: actions to complete 90, 60, and 30 days before a regulatory exam
Build the plan in three windows: scope, inventory, and gap assessment at 90 days; remediation, validation, and evidence collection at 60 days; evidence packaging, walkthrough rehearsals, and change control at 30 days. Each window needs named owners, dated deliverables, and exit criteria.
Start from what examiners need to see
The plan sequences inventory, gap remediation, evidence packaging, and walkthrough rehearsals for AI models, vendor AI, and agents before a bank exam. Work backward from the exam date, and give each window named owners, dated deliverables, and exit criteria.
| Window | Focus |
|---|---|
| 90 days out | Scope, AI inventory, risk tiering, and gap assessment |
| 60 days out | Remediation, control validation, access and policy review, evidence collection |
| 30 days out | Evidence packaging, walkthrough rehearsals, and change controls |
90 days out: scope, inventory, tiering, and gap assessment
- Set the exam scope.
- Build the AI inventory, covering models, vendor AI, and agents.
- Risk-tier each system.
- Run a gap assessment and record each gap with severity, owner, and target date.
60 days out: remediate, validate, review, and collect
- Remediate the gaps identified in the assessment.
- Validate controls, with validation records linked to each system's tier.
- Review policies and access, including agent identity and permissions.
- Collect evidence, including logs, incident records, and change records.
30 days out: package, rehearse, and control change
- Package evidence into an indexed set.
- Rehearse examiner walkthroughs.
- Freeze or formally control changes.
Using runtime logs and agent records as evidence
Where AI agents are in scope, the evidence set includes agent identity, permission, and access review records, along with logs, incident records, and change records. Runtime enforcement, agent identity, and audit logging can support the evidence a readiness plan requires.
Artifact and owner checklist
| Artifact | Owner |
|---|---|
| AI inventory with tiers, covering models, vendor AI, and agents | AI governance or model risk lead |
| Gap register with severity, owner, and target date | Readiness lead in risk |
| Validation records linked to each system's tier | Model validation |
| Third-party due diligence and monitoring files for AI services | Third-party risk management |
| Agent identity, permission, and access review records | Information security |
| Indexed evidence set with logs, incident records, and change records | Compliance with technology |