How SWT3 Witness Anchors provide examiner-ready evidence for model risk management under the revised interagency guidance
Supersession Notice (April 17, 2026)
SR 26-2 and OCC Bulletin 2026-13 jointly supersede SR 11-7, SR 21-8, OCC Bulletin 2011-12, OCC Bulletin 2021-19, OCC Bulletin 1997-24, and the Comptroller's Handbook MRM booklet. The revised guidance was issued jointly by the Federal Reserve, OCC, and FDIC.
Key change for AI practitioners: SR 26-2 explicitly states that "generative AI and agentic AI models are novel and rapidly evolving. As such, they are not within the scope of this guidance." A future request for information (RFI) addressing AI-specific model risk management is planned but not yet issued.
What this means: Traditional quantitative models (credit scoring, fraud detection, capital calculation) remain governed by SR 26-2. GenAI and agentic AI systems fall into a regulatory gap. The SWT3 procedures documented below apply to both traditional MRM and the emerging AI governance space. Institutions deploying AI models should not wait for the RFI to establish governance practices.
Who this is for: Model Risk Management Officers, Chief Risk Officers, internal auditors, and OCC/Federal Reserve examiners evaluating model governance frameworks. If your institution uses AI or statistical models for credit, fraud, pricing, or capital decisions, this document maps SWT3 procedures to SR 26-2 (and its predecessor SR 11-7) examination requirements.
SR 26-2 modernizes model risk management expectations while narrowing the definition of "model" to exclude GenAI and agentic AI. Key changes from SR 11-7:
| Area | SR 11-7 (2011) | SR 26-2 (2026) |
|---|---|---|
| Model definition | Broad: "quantitative method, system, or approach" | Narrowed: adds qualifier "complex" and excludes simple arithmetic and deterministic rule-based processes |
| GenAI / Agentic AI | Not addressed | Explicitly excluded from scope; future RFI planned |
| Proportionality | One-size-fits-all | Tailored to institution size, complexity, and model footprint; most directly applies to institutions with >$30B assets |
| Issuing agencies | Fed + OCC | Fed + OCC + FDIC (joint interagency) |
| Binding status | Supervisory guidance | Explicitly non-binding; non-compliance alone will not trigger supervisory criticism |
| Supersedes | N/A | SR 11-7, SR 21-8, OCC 2011-12, OCC 2021-19, OCC 1997-24, Comptroller's Handbook MRM |
SWT3 defines 5 procedures for model risk management. Each procedure produces a tamper-evident Witness Anchor that proves a specific MRM activity occurred, when it occurred, and what the outcome was. These procedures apply to both traditional quantitative models governed by SR 26-2 and AI/ML models that fall outside its explicit scope.
| Procedure | What it proves | SR 11-7 Section | Examination Focus |
|---|---|---|---|
| FIN-GOV.1 | Governance committee reviewed and approved the model | Section III | Committee minutes, quorum, approval votes |
| FIN-MRM.1 | Model registered, version matches approved inventory | Section V | Inventory completeness, version lineage |
| FIN-VAL.1 | Independent validator performed effective challenge | Section VI | Validator independence, sign-off records |
| FIN-MON.1 | Performance metric within threshold, no drift | Section VII | PSI/CSI thresholds, alerting frequency |
| FIN-OUT.1 | Back-test completed, results within tolerance | Section VIII | Sample adequacy, prediction accuracy |
SR 11-7 requires: The board and senior management should ensure model risk management is part of the overall risk framework. A governance structure with clear roles and responsibilities should be established.
How SWT3 addresses it: When the Model Risk Governance Committee votes to approve a model, the vote is anchored with three factors: quorum requirement (factor_a), actual vote count (factor_b), and approval decision (factor_c). The anchor proves a properly constituted committee made a documented decision on a specific date.
The SWT3 anchor for FIN-GOV.1. Factor values confirm quorum was met and the vote was recorded. Cross-reference with your committee charter to confirm the quorum threshold matches governance policy. The anchor is independently verifiable with no vendor dependency.
SR 11-7 requires: A firm-wide model inventory should be maintained with version lineage, purpose, limitations, and risk tier for each model.
How SWT3 addresses it: Each time a model is deployed or updated, the SDK records whether the deployed model hash matches the approved version. PASS means production matches inventory. FAIL means a mismatch was detected, which could indicate an unauthorized model change.
The SWT3 ledger filtered by FIN-MRM.1 for your model. A continuous chain of PASS verdicts demonstrates the model in production has always matched the approved inventory. Any FAIL creates an auditable record of when a mismatch was detected.
SR 11-7 requires: Validation should be conducted by qualified staff independent of model development. "Effective challenge" involves critical analysis by objective, informed parties.
How SWT3 addresses it: When an independent validator signs off, the sign-off is anchored. Factor_b confirms the validator signed. Factor_c records days since last validation for staleness tracking. The anchor proves effective challenge occurred on a specific date.
The SWT3 anchor for FIN-VAL.1 along with the validator's identity from ledger metadata. The examiner confirms: validation occurred, when it occurred, and the record hasn't been altered. Factor_c provides a built-in staleness indicator to compare against your validation frequency policy.
SR 11-7 requires: Models should be subject to ongoing monitoring to confirm they continue to perform as expected. Deterioration should trigger review.
How SWT3 addresses it: The SDK records the performance metric (PSI, CSI, AUC, Gini) against its defined threshold at each monitoring interval. PASS means within bounds. FAIL means the threshold was breached, creating a tamper-evident record of when drift was detected.
The time series of FIN-MON.1 anchors. The examiner sees: monitoring frequency, consecutive PASS verdicts showing stability, and if drift occurred, exactly when. The cryptographic integrity layer binds all monitoring anchors into a single tamper-evident verification artifact.
SR 11-7 requires: Outcomes analysis compares model outputs to actual outcomes. This should be performed regularly with sufficient sample sizes.
How SWT3 addresses it: Each back-testing cycle is anchored with: required sample size (factor_a), actual sample size (factor_b), and whether results were within tolerance (factor_c). PASS means adequate data and predictions aligned with reality.
The SWT3 anchor for FIN-OUT.1 from the most recent cycle. Factor_a vs factor_b proves sample adequacy. Factor_c proves tolerance was evaluated. The timestamp proves when the analysis was conducted. Compare frequency against your MRM policy.
SR 26-2 creates a documented governance gap for financial institutions deploying generative AI and agentic AI. The guidance explicitly states these technologies are "not within the scope" while acknowledging they present novel risks. Institutions cannot wait for the planned RFI to establish governance practices.
The following AI-namespace procedures provide witness evidence for GenAI/agentic governance activities that SR 26-2 does not cover but examiners will increasingly expect:
| Procedure | What it proves | Gap addressed |
|---|---|---|
| AI-MDL.1 | Model card registered with version, architecture, and intended use documented | GenAI model inventory (SR 26-2 Section V equivalent for AI) |
| AI-DRIFT.1 | Output drift detected or confirmed absent at monitoring interval | GenAI performance monitoring (SR 26-2 Section VII equivalent) |
| AI-GRD.1 | Safety guardrails active and tested | GenAI output controls (no SR 26-2 parallel) |
| AI-FAIR.1 | Bias evaluation performed, results within threshold | Fair lending / disparate impact for AI scoring models |
| AI-GOV.1 | AI governance policy reviewed and approved | Board-level AI oversight (SR 26-2 Section III equivalent) |
| AI-EXPL.1 | Explainability method applied, output interpretable | Model transparency for complex AI (effective challenge) |
| AI-HITL.1 | Human-in-the-loop review performed on high-risk decision | Human oversight for automated decisions |
These procedures produce the same tamper-evident Witness Anchors as the FIN procedures above. An institution can present a unified evidence chain covering both traditional models (FIN-*) and AI models (AI-*) to demonstrate comprehensive governance regardless of the regulatory gap.
| Examiner question | Where to look |
|---|---|
| "Show me your model governance documentation" | FIN-GOV.1 anchors. Each represents a committee vote with quorum and approval status. |
| "Is this model in your inventory?" | FIN-MRM.1 anchors. PASS = hash matches approved version. FAIL = mismatch detected. |
| "Who validated this model?" | FIN-VAL.1 anchors + ledger metadata for validator identity. |
| "How often do you monitor performance?" | FIN-MON.1 anchor frequency. Daily anchors = daily monitoring. |
| "Has this model drifted?" | FIN-MON.1 factor_c values. 0 = no drift. 1 = threshold breached. |
| "When was the last back-test?" | FIN-OUT.1 most recent anchor timestamp. |
| "Can I verify this independently?" | Yes. Public verifier or offline SHA-256 formula. No vendor dependency. |
| "How do I know records weren't altered?" | Each anchor is a SHA-256 fingerprint. A periodic integrity rollup binds all anchors into a tamper-evident digest. Tampering breaks the chain. |
| "Where is the data stored?" | Clearing Level 2+: factors only, no model details on TeNova side. |
When documenting SWT3 in your Model Risk Management Policy, consider language similar to the following: