Orient banking, Farm Credit, insurance, and securities teams across supervisory guidance, examination practice, and the recordkeeping obligations that already attach to AI-assisted work.
The United States has no AI Act. What US financial institutions face instead is a layered mix of supervisory guidance, examination practice, and hard recordkeeping obligations attached to existing statutes. Four supervised populations answer to different agencies with different instruments, so start by identifying which one owns the workflow. This page describes the landscape as of August 2026 and maps it to shipped AxonFlow capabilities. It is not legal advice, and AxonFlow does not certify compliance with any instrument named here.
Model risk discipline under SR 26-2 and OCC Bulletin 2026-13, interagency third-party risk management, GLBA Safeguards activity logging, and FFIEC BSA/AML expectations for AI-assisted monitoring.
Read the banking mappingExamined by the FCA rather than the federal banking agencies: Exam Manual EM-31.1 model risk discipline, the 12 CFR Part 609 cyber rule, and NIST AI RMF as the named external framework.
Read the Farm Credit mappingThe NAIC Model Bulletin on the use of AI systems, the AI Risk Evaluation Supplement questionnaire, NYDFS Circular Letter No. 7, Colorado Regulation 10-1-1, and Texas TDI Bulletin B-0003-26.
Read the insurance mappingNo AI rule after the June 2025 withdrawal of the predictive data analytics proposal. Examination priorities, sweep letters, existing antifraud law, and books-and-records obligations that attach to AI outputs and agent actions.
Read the securities mappingFederal examiners probe AI use in routine examinations without an AI rulebook to cite. These are the questions that recur, and they are answered with evidence rather than attestations. The same set is a workable self-assessment for any of the four segments above.
| Examiner question | What AxonFlow contributes | What stays with you |
|---|---|---|
| Where is AI in use, and through what systems? | Governed agents, policies, and LLM providers are registered and observable on every path routed through AxonFlow. | The institution-wide AI inventory. AxonFlow reports what it governs and does not claim network-wide discovery of ungoverned AI. |
| Can you reconstruct an individual AI decision? | Per-decision records carrying the decision identifier, evaluated policies, verdict, identity attribution, and timestamps across the agent, orchestrator, connector, and workflow layers. | Correlation with the downstream core banking, policy administration, or order-management event. |
| Who oversees high-risk actions, and can you prove it? | Human-in-the-loop approval gates pause configured actions for review, and the approval record carries approver identity, justification, and timing. Creating approval entries requires a paid licence (Professional or higher); below that, a require_approval policy still holds the step but creates no queue entry. | Reviewer authority, escalation paths, separation of duties, and the risk appetite that decides which actions need a human. |
| Can an AI system be shut down? | Circuit-breaker and kill-switch controls (Enterprise edition) halt governed paths on failure thresholds and disable AI systems, and policy changes take effect at the enforcement point without redeploying agents. | Activation authority, recovery rehearsal, and business continuity. |
| What data can the AI reach, and what was denied? | Policy-scoped data access on governed paths, with denials recorded in the same audit surface as approvals so denied-attempt evidence is producible. | Data classification, lawful basis, and the access model outside governed paths. |
| How is customer information protected in prompts and tool calls? | PII detection with configurable block, redact, warn, and log actions, including US patterns such as Social Security numbers with format validation and US bank routing and account numbers. | The wider GLBA Safeguards program, including service-provider oversight and board reporting. |
| How do you evidence any of the above to us? | Evidence Export (not included on Community) produces structured, time-bounded packages over audit logs, workflow steps, and approval records. | Presenting that evidence inside your own examination and model risk processes. |
US evidence is produced through the framework-agnostic Evidence Export and the audit APIs. The regulator-specific compliance report feature currently covers the EU AI Act, SEBI, RBI FREE-AI, MAS FEAT, and the Indonesian frameworks (OJK, Bank Indonesia, and UU PDP), and no US regulator is among them.
On April 17, 2026 the Federal Reserve, OCC, and FDIC issued revised model risk management guidance (SR 26-2, OCC Bulletin 2026-13, and FDIC FIL-15-2026), superseding SR 11-7 and OCC 2011-12. The revised guidance is stated to be most relevant to banking organizations above $30 billion in assets, and it retains the pillars examiners have applied for fifteen years. One scope fact matters for AI teams: the revised guidance narrows the formal model definition and excludes generative and agentic AI from its scope, while the agencies indicate that the underlying risk-management principles still apply.
| Requirement | Instrument | AxonFlow contribution |
|---|---|---|
| Model risk discipline for AI systems: inventory support, monitoring evidence, and documentation a knowledgeable third party can follow | SR 26-2 and OCC 2026-13, successors to SR 11-7 | Per-decision audit records, approval history, and monitoring evidence that a model risk team folds into its own documentation. AxonFlow does not perform model validation or effective challenge. |
| Life-cycle governance of third-party AI relationships | Interagency Guidance on Third-Party Relationships (2023) | LLM provider inventory and routing controls, per-provider usage and decision records, and self-hosted deployment inside the institution's own environment. |
| Monitor and log the activity of authorized users on systems holding customer information | GLBA Safeguards and the Interagency Guidelines | Identity-attributed records of every governed model and tool call, with PII detection and configurable redaction before model or connector access. |
| Governance of AI-assisted financial crime monitoring, with the records that show what triggered or suppressed an alert | FFIEC BSA/AML Examination Manual and FinCEN guidance | The FinCrime Policy Pack rules, part of the Fraud and Risk Add-on, are authored from public FFIEC examination guidance, FinCEN advisories, OFAC sanctions programs, and 31 CFR 1010.311 thresholds, and detections land in the standard audit surface. |
| Support for specific adverse-action reasons where AI touches credit | ECOA and Regulation B, 12 CFR 1002.9 | Per-decision records identify the policies evaluated and the actions gated, supporting the institution's own adverse-action process. AxonFlow does not generate adverse-action notices and does not perform fair-lending or statistical fairness testing. |
| Map AI governance evidence to the voluntary frameworks examiners reference | NIST AI RMF and the Treasury-announced, industry-built Financial Services AI Risk Management Framework | Policy enforcement, approval, and audit evidence align with the Govern, Map, Measure, and Manage functions. The final mapping remains with the institution's risk team. |
Farm Credit System lenders are supervised by the Farm Credit Administration, not the federal banking agencies, and the FCA's expectations for AI are concrete. The banking mapping above applies, with the differences below.
A risk-tiered inventory including vendor and System-shared models, independent validation and effective challenge, a change-control log recording when, what, who, and approver, and board reporting. Model risk itself is not outsourced. Where it is unclear whether a tool is a model, EM-31.1's direction is to treat it as one, which pulls AI tools into scope. The change-approval evidence that AxonFlow's approval and policy-versioning records produce matches that when, what, who, and approver form.
Effective January 2025: a board-approved written cyber risk program reviewed annually, risk assessment of all vendors, independent control testing, incident notification to the FCA within 36 hours, and quarterly board reporting. Audit and evidence surfaces support the quarterly rather than annual reporting cadence.
The FCA names the NIST AI Risk Management Framework and its Generative AI Profile directly on its AI guidance page, so governance evidence organized around the Govern, Map, Measure, and Manage functions is what examiners are pointed to.
The FCA's FY2026 oversight plan includes evaluating a sample of institutions on their use of AI in risk management and operations, which makes the standing examiner question set above an immediate self-assessment rather than a future one.
Insurance AI governance is set by state regulators working from NAIC instruments, with several states adding their own. The instruments converge on an AI systems program with board accountability, an inventory with risk classification, third-party accountability, and lifecycle documentation producible to a regulator.
| Requirement | Instrument | AxonFlow contribution |
|---|---|---|
| AI system inventory with change tracking and approvals, including retired models | NAIC Model Bulletin on the Use of AI Systems by Insurers, adopted in 25 jurisdictions as of August 2026; NYDFS Insurance Circular Letter No. 7 (2024) | Policy versioning and approval records carrying approver identity and timestamps. |
| Lifecycle documentation producible to a regulator | NAIC Model Bulletin | Per-decision audit records and time-bounded Evidence Export packages. |
| Disclosure of the information an adverse decision relied on, and its source | NYDFS Circular Letter No. 7 | Decision records enumerate the policies evaluated and the data categories acted on for governed paths. A proprietary vendor algorithm is not an excuse under CL7, so vendor-side reasoning remains the insurer's to obtain. |
| Examiner questionnaire evidence: quantified AI usage, governance framework, per-high-risk-system detail, data inputs and sources | NAIC AI Risk Evaluation Supplement, piloting in 12 states as of August 2026, running through September 2026, with adoption targeted for late 2026 | Decision volumes, governance policies, per-system audit history, and override records provide source evidence for the supplement's exhibits. |
| Governance and risk management frameworks for external consumer data and information sources, with annual compliance reports naming accountable individuals | Colorado Regulation 10-1-1, life insurers since 2023 and expanded to private-passenger auto and health insurers effective October 2025 | Governed connector and data-access records support the framework's evidence, and approval records name the humans who authorized changes. AxonFlow does not perform the quantitative fairness testing these frameworks also require. |
| Quantitative fairness analysis before deployment and at least annually | NYDFS Circular Letter No. 7 and Colorado Regulation 10-1-1 | Not an AxonFlow capability. Fairness and bias testing remains the insurer's own actuarial and data-science work. AxonFlow records complement that evidence, they do not produce it. |
Texas TDI Bulletin B-0003-26 (June 2026) adds that decisions made or supported by AI must comply with unfair-discrimination law, that governance extends to third parties, and that TDI examinations may probe AI governance and human oversight.
The SEC withdrew its predictive data analytics proposal in June 2025, so there is no AI rule. What operates instead is examination practice and existing law.
AxonFlow's fit here is the same evidence spine: identity-attributed records of what AI produced and did, approval dispositions for outputs that constitute recommendations or actions, and export packages in reviewable formats. The compliant archival system, whether WORM or the audit-trail alternative, remains the firm's own.
The binding US obligations that reach AI work are mostly recordkeeping obligations. They survive shifts in federal AI policy, which makes them the safest thing to design a retention configuration against.
| Records | Instrument | Retention |
|---|---|---|
| Adverse-action records | Regulation B, 12 CFR 1002.12(b) | 25 months |
| Suspicious activity reports and supporting documentation | BSA, 31 CFR 1020.320(d) | 5 years |
| Books and records | SEC Rule 17a-4 and Rule 204-2 | 5 to 6 years; WORM or the 2022 amendments' audit-trail alternative for broker-dealers |
| Cybersecurity records | NYDFS, 23 NYCRR 500.06 | 5 years for transaction-reconstruction records; 3 years for cybersecurity-event audit trails |
| Automated decision-making technology records | Colorado SB 26-189, effective January 2027 | 3 years |
Audit retention is fixed on Community (3 days) and Evaluation (14 days), and is configurable up to 10 years on the paid tiers. Neither Community nor Evaluation can meet any obligation in the table above, so plan on a paid-tier retention configuration or an archive outside AxonFlow, aligned with the strictest instrument that applies to the workflow.
AxonFlow is source-available under BSL 1.1, so a model risk or security team can inspect the enforcement, detection, and audit implementations directly rather than relying on a vendor attestation. Legal and compliance teams must still validate the complete system against current law, supervisory guidance, contractual obligations, and deployment facts.
Use the Evaluation license to assess policy enforcement, human review, evidence records, and deployment boundaries before a regulated rollout.