Banking AI agent · Programme governance
AML Programme-Governance & Independent-Testing AI Agent
Assemble the programme evidence a board approves and a named independent tester opines on, reconcile the risk assessment to the controls, and track each order deliverable against the day count it carries.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Board approves
- Minute-backed
- After launch
- Agent Care
What this agent does
Maintains the evidence, not the approval
In
- 01
A pillar is tested, and the agent opens one evidence set from the policy, training, monitoring and filing systems.
- 02
A legal-entity file is opened, and CDD at 31 CFR 1010.230 still applies — only foreign companies now file BOI.
Reason
- 03
A minute entry is drafted — the OCC asks for approval reflected in the minutes, the Fed and FDIC for it noted.
- 04
A risk is retired, and the agent shows which controls still map to it and which risks are now left uncovered.
- 05
A risk profile shifts, and the scope reopens — no rule sets a testing frequency, and change is the trigger.
Decide
- 06
A finding is raised, and it is logged against the element it hit, its owner and the evidence closure will need.
- 07
A training population is defined, and contractors, mid-period joiners and acquired lines are named in or out.
Out
- 08
An order article falls due, and the agent tracks the day count and whether no supervisory objection was sought.
- 09
Execute write actions only inside the approval boundaries agreed during implementation.
Product statement
The board approves and a named tester opines; if a signed opinion rested on a pack the agent mis-assembled, the next examination re-tests that period.
Example workflow
One element, evidence to opinion
- 1Testing cycle openedBoard calendar, examination findings, order articles or a change in the risk profile
- 2Evidence assembledProgramme version, risk assessment, control mapping, training records and issue log, each with its source
- 3Pack draftedElement coverage, evidence gaps, open findings and confidence
- 4Controls appliedSource-currency checks, control-to-risk reconciliation, scope-coverage rules and confidence threshold
No human action required
Stages 1 to 4 run unaided, and nothing is approved or opined at any of them — the agent is assembling, and the officer's lane opens at the confidence gate.
Goes to the BSA officer to review.
Adds a second compliance read first.
Officer review
The pack is held with its evidence, its gaps and the confidence.
Accept · Amend · Send to second review
Accepted — goes to the board and the tester
- 6Programme record updatedOnly where write access and approval policy allow it
- 7Outcome evaluatedGaps the tester found, officer amendments, issues reopened and examination findings
Amendments
Every officer amendment is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Approving the programme — the board's own act.
- Signing the scoped, dated testing opinion.
- Accepting residual risk on any element.
- Closing an MRA or a consent-order article.
Automation boundary
Agent acts unaided
- ✓Assemble the programme record and stamp what was current when.
- ✓Reconcile the risk assessment to the controls mapped to it for the named owner.
- ✓Test the evidence set against the configured element rules.
- ✓Flag what the pack is missing, and hold it for the officer.
Any write happens inside the boundaries agreed at implementation, never ahead of the board.
- Judging whether a tester is independent.
- Setting the scope of the independent test.
- Certifying an order deliverable to the OCC.
- Changes to element, evidence or escalation rules.
Example output
One programme element, annotated
The examinable artefact is the minute naming the version approved — this is that layer.
Programme evidence · single element
Illustrative example- Element
- Internal controls
- Evidence assembled
- Control-to-risk mapping reconciled to the current assessment
- Version
- v4.2, 12 Mar 2026
- Rule and record
- 12 CFR 21.21(c)(1) · OCC
- Confidence
- 91%
- Approval record
- Minute entry, board of directors
As received
Taken from the programme record and the board minute — nothing on this side is written by the agent.
Evidence held
- Approved programme v4.2
- Risk-assessment extract
- Training completion log
Why this record
The minute names the version the board approved and a person still decides.
What the score decides
Below the configured threshold the pack picks up a second compliance read first.
Action
AcceptAmendSend to second reviewValue
Where AI adds value
The same four claims, placed at the point in the workflow where each one applies.
Where the value lands
Value 01 – 04
Every element
From the programme record
Assemble from the record
Draw on the programme record, the element rules and the day counts a consent order sets.
The minutes are the evidence
Most programme tooling holds the document and not the minute — the half that proves nothing.
Send review to the weak evidence
Unsupported closures and low-confidence packs are marked, so the officer's read starts where the gaps sit.
Build an evidence trail
The element, the testing it was scoped for and the tester who opined stay on the programme.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
GRC and policy
- Archer
- MetricStream
- OneTrust
- ServiceNow IRM
Audit management
- AuditBoard
- Workiva
- TeamMate
- Galvanize
AML systems
- Actimize
- Verafin
- Oracle FCCM
- Hawk
Agent
AML programme governance & independent testing
- Reads the record
- Assembles evidence
- Holds for the officer
Board and training
- Diligent
- Boardvantage
- Cornerstone
- Docebo
Observability & evaluation
- OpenTelemetry
- Langfuse
- Supported monitoring/evaluation sources
Integration availability depends on the client's existing systems and API access.
Agent controls
Six layers between the model and the programme
Six controls, every one holding the next. What none of them holds is named in the map below.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether a pack may stand
- L3
- Puts the acceptance in an officer's hands
- L4 – L5
- Keep the element and the testing behind it
- L6
- Steps back to evidence listing when signals degrade
How Nestack evaluates it
Evaluate the evidence workflow — not only the pack the board reads.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the pack the board and the tester read
- E1Final-output evaluationDid every item in the pack match its source system?
- E2Step-level evaluationDid the agent use the right programme version and element rules?
- E3Tool evaluationDid it read the correct element and the correct source system?
- E4Confidence calibrationDo low-confidence packs actually attract more officer amendments?
- E5Slice evaluationHow does performance change across specific programme elements?
- E6Business outcomeHow many packs needed an officer amendment or a later correction?
Floor — the outcome the bank answers for
Failure modes
Where each failure originates in the agent
Seven failure modes, placed at the stage each one originates.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- OP-03
Superseded version used
Evidence pulled from a version the board never approved.
Stage gathers
Programme versions, risk assessment, training, issues
02 · Reasoning
2 modes
- OP-04
Orphaned control mapping
A control still maps to a risk the assessment retired.
- OP-06
Scope inherited unchanged
Testing scope carries over though the bank's risk moved.
Stage proposes
Element coverage, evidence gaps and confidence
03 · Tool / write
2 modes
- OP-02
Closure without evidence
An issue closes on an attested date with nothing attached.
- OP-05
Population left undefined
Training completion is reported over an unstated group.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- OP-01
Approval not minuted
A version reaches the board with no minute entry drafted.
Stage returns
The pack the officer accepts and the board approves
05 · Change / Version
1 mode
- OP-07
Silent element drift
A model or rule change narrows what the agent flags.
Stage tracks
Model, prompt, element rules and programme config
- Sev-1 · pack relied on outside the boundary
- Sev-2 · a stale version reaches the board
- Sev-3 · source degrades, pack routes to review
Affected slices
An evidence gap is never evenly spread
A programme-wide evidence figure is carried by the pillars that hold the most documents, while the thin ones set the examination. Nestack reports the evidence-gap rate by pillar, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Board minute record | 7.6% | 3.6× | Review | |
| Issue and MRA closure | 5.5% | 2.6× | Review | |
| Risk-assessment currency | 3.7% | 1.8× | Watch | |
| Internal-controls testing | 1.6% | 0.8× | Normal |
Bar: evidence-gap-rate lift vs. internal-controls baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
Each cycle closes with a new evidence case
A cycle is done when the undocumented approval has become a case the next release must close. That suite is what the next test scoped is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Evidence gaps rise in one programme element.
- 02
Diagnose
The programme approval nobody could find in a board minute is traced back to one cause — a version number the corporate secretary was never given.
- 03
Improve
The fix gets a number, and the findings that raised it are held against it.
- 04
Verify
No release while a touched evidence case is outstanding; it reruns until green.
- 05
Learn
The case stays for good, and the elements it spans are recorded against it.
Learn → Detect
The return edge. The next detection runs against a suite one evidence case longer.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, sources, evidence workflow, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Programme workflow discovery and boundary setting.Week 1
- 02GRC, audit and training-source assessment.Week 1, Week 2
- 03Element, evidence and escalation-rule mapping.Week 1, Week 2
- 04Programme-record ingestion and mapping.Week 2
- 05Control-to-risk reconciliation logic.Week 2, Week 3
- 06Confidence scoring and escalation routing.Week 3
- 07Officer review workflow.Week 3, Week 4
- 08GRC and audit-platform integration.Week 5
- 09Evidence and remediation cases.Week 4, Week 5
- 10Guardrails and opinion controls.Week 4
- 11Programme-trail instrumentation.Week 5, Week 6
- 12Deployment, documentation and Agent Care handover.Week 6
12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallel
Final scope and sequence confirmed in discovery
Engagement tiers
What each tier includes
Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.
| Capability✓ in scope · — not at this tier | PilotOne charter, one programme | ProductionProduction GRC systems | AdvancedMultiple charters / entities |
|---|---|---|---|
| Introduced at Pilot | |||
| Assembly to your record and elements | ✓ | ✓ | ✓ |
| Officer review | ✓ | ✓ | ✓ |
| Evidence-coverage baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by pillar | — | ✓ | ✓ |
| Review workflow in your systems | — | ✓ | ✓ |
| Approved write-back | — | ✓ | ✓ |
| GRC-platform integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-agency element rules | — | — | ✓ |
| Multi-stage board approvals | — | — | ✓ |
| High issue volume | — | — | ✓ |
| Multi-charter programme controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on integrations, workflow complexity, transaction volume, approval controls and deployment requirements.
- Separate from build
- Build pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.
What we need from you
What you bring, and what we build with it
Each input maps to a piece of build scope and a week in the delivery timeline.
You bring
We build with it
01Your programme document and element structure
→Programme-record ingestion and evidence mapping
Week 102Representative past testing cycles
→Evidence baseline, source binding and coverage mapping
Week 203Your risk assessment and control inventory
→Element, evidence and escalation-rule mapping
Week 104Access to relevant APIs, feeds or exports
→GRC, audit and training-source assessment, then integration setup
Week 205Programmes you would not want tested
→Finding cases and failure-mode testing
Week 406What no opinion may cover
→Confidence scoring, escalation routing, guardrails and approval controls
Week 307Named officers, and a tester reporting to the board
→Officer review workflow, then pilot and production validation
Weeks 5–6
Nothing else is required
Deployment, documentation and Agent Care handover are ours.
Delivery timeline
Four phases across six weeks
The weeks are taken from the board calendar, so the fifth band carries evaluation and the pilot.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Programme workflow discovery, element mapping and the automation boundary
W2Source integration and the evidence baseline
W3Evidence workflow, reconciliation logic and review controls
W4Evaluation suite, coverage checks and failure-mode testing
W5GRC integration, pilot elements and targeted corrections
W6One testing cycle run under the AML officer, then Agent Care handover
- Reading the band
- A bar covers the weeks its work is named in, and nothing else. The week 5 overlap is real, not padding.
- At the end of W6
- Once the cycle validates, monitoring transfers to Agent Care.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Banking AI agent
Build a programme-governance agent around the minute that proves approval.
Show us your programme and your issue log. Your corporate secretary holds the minute an examiner asks for first, and we build to the rule in force — the 2024 NPRM is withdrawn, no successor final.
- Typical delivery
- 4–6 weeks
- Deployment
- Your GRC stack
- Board approves
- Minute-backed
- After launch
- Agent Care available
Nestack Agents · AML programme governance
AGT-BK-19 · Agent Care available after launch