Skip to main content
    Kriv AI

    Financial Services Governance

    Model Risk Management for AI: From SR 11-7 to SR 26-2

    SR 26-2 excludes generative and agentic AI models from its scope. A bank's own model risk management and governance should still cover them.

    context

    Current status

    SR 11-7 is no longer the current model risk management guidance. The Federal Reserve, OCC and FDIC issued SR 26-2 on April 17, 2026, and it supersedes SR 11-7 and SR 21-8, the 2021 statement on model risk for Bank Secrecy Act and anti-money laundering systems. The Federal Reserve's letter says SR 26-2 is expected to be most relevant to banking organizations with over $30 billion in total assets that the Federal Reserve regulates. SR 26-2 says generative AI and agentic AI models are not within its scope. For tools it does not cover, a bank's own risk management and governance practices should guide the controls.

    context

    SR 11-7 vs SR 26-2: what changed

    SR 26-2, issued on April 17, 2026 by the Federal Reserve, the OCC and the FDIC, supersedes and replaces SR 11-7. Read at source, SR 26-2 defines a model as a complex quantitative method that applies statistical, economic or financial theories to process input data into quantitative estimates, and it leaves out simple spreadsheet arithmetic and deterministic rule-based processes. It treats model materiality, which combines a model's exposure and its purpose, as an important consideration. It says it does not set forth enforceable standards, and it is expected to be most relevant to banking organizations with over 30 billion dollars in total assets. We read SR 26-2 as written and do not compare it line by line with the 2011 text here.

    SR 11 7: what to cite now

    SR 11 7 is the Federal Reserve's 2011 model risk letter, usually written SR 11-7, and it has been superseded. Cite SR 26-2 of April 17, 2026 for model risk management. For national banks, federal savings associations and federal branches and agencies, the OCC issued Bulletin 2026-13, which rescinds OCC Bulletin 2011-12. SR 26-2 also supersedes SR 21-8, the 2021 statement on systems supporting Bank Secrecy Act and anti-money laundering compliance, and OCC Bulletin 2026-13 rescinds OCC Bulletin 2021-19. If a policy, contract or exam response of yours still cites SR 11-7, update the citation and check that the text around it still fits.

    SR 11-7 guidance on model risk management: what SR 26-2 now says

    The SR 11-7 guidance on model risk management, issued in 2011, has been replaced by SR 26-2, the supervisory guidance on model risk management that the Federal Reserve, the FDIC and the OCC issued on April 17, 2026. SR 26-2 says it highlights sound principles for effective model risk management and that each banking organization is responsible for adopting practices that fit its own risks. Its parts cover model development and use, validation and monitoring, governance and controls, and vendor and other third-party products. It says it does not set forth enforceable standards or prescriptive requirements. If you were looking for SR 11-7 guidelines or principles, read SR 26-2 instead.

    context

    What SR 26-2 Covers and What It Leaves Out

    Why Generative AI and LLMs Raise New Model Risk Questions

    SR 11-7, the Federal Reserve's 2011 model risk management guidance, predates large language models and does not address them explicitly. A large language model or an agentic AI system behaves differently from a traditional model: its behavior can vary by prompt, its outputs are generative rather than a single predicted value, and a single deployment can span dozens of use cases rather than one model, one purpose.

    Generative AI Model Risk Management Under SR 26-2

    SR 26-2 does not cover generative AI model risk management: its footnote 3 says generative AI and agentic AI models are novel and rapidly evolving and are not within the scope of the guidance, and that a banking organization's risk management and governance practices should guide the controls for tools it does not cover. SR 26-2 does not extend to generative and agentic AI: its own footnote says these models are novel and rapidly evolving and are not within the scope of the guidance, and the guidance does not set forth enforceable standards. SR 26-2 footnote 1 says: "See 12 CFR Part 4, Subpart F, Appendix A (OCC); 12 CFR Part 262, Appendix A (Board); 12 CFR Part 302, Appendix A (FDIC). However, supervisory action may result for any violations of law or unsafe or unsound practices stemming from insufficient management of model risk." Footnote 3 adds that a banking organization's risk management and governance practices should guide the determination of appropriate governance and controls for tools it does not cover. In Kriv AI's practice, those controls can focus on output quality sampling, guardrail testing, and use-case-specific risk tiering rather than a single accuracy number.

    framework

    Three Lines of Defense, Reinterpreted for AI

    In Kriv AI's practice, the first line (the business unit deploying the model), second line (independent risk and compliance oversight), and third line (internal audit) structure still applies, but AI adds new questions to each line: who owns a prompt-engineering change to a production LLM, who validates an agent's tool-use permissions, and who audits a vendor-supplied foundation model your team doesn't control the training data for.

    validation

    Model Validation for LLMs and Agentic AI

    Why Validation Looks Different for Generative Models

    Generative models produce open-ended output rather than a single predicted value. Validation instead can use structured output sampling against a rubric, red-teaming for guardrail failures, and ongoing drift monitoring on output quality rather than a single input-output accuracy check.

    inventory

    Model Inventory and Risk Tiering

    Every AI system in production, including third-party vendor tools with an embedded LLM, needs to be in a model inventory with a risk tier assigned based on the decision it influences: a customer-facing chatbot answering general questions sits in a different tier than an agent that can initiate a wire transfer or approve a loan.

    governance

    Governance, Roles, and Board Reporting

    SR 26-2 does not cover generative and agentic AI, so board and senior-management reporting on AI model risk comes from the bank's own governance. A bank or asset manager should have a defined owner for AI governance, not an informal arrangement where whoever deployed the model also validates it. Model risk is one part of a wider AI compliance framework.

    SR 11-7 model governance: roles, inventory and controls under SR 26-2

    For SR 11-7 model governance questions, the current text is SR 26-2, which says all parts of the model lifecycle benefit from governance supported by clear policies and effective controls. It says policies define risk management expectations and set a framework for judging how large model risk is, and that procedures support the policies. It asks for clear roles and well-defined accountability, including for conflicts of interest such as misaligned incentives between model development and validation groups. It describes a model inventory with enough information to understand model risk individually and in aggregate, and it says adequate documentation supports effective model risk management. Which controls your bank applies is for its own model risk team and its supervisors.

    governance

    Compliance today

    SR 11-7 compliance now means meeting SR 26-2, because SR 26-2 replaced the 2011 guidance. SR 26-2 narrows the definition of a model to a complex quantitative method, and it leaves out simple arithmetic and deterministic rule-based processes. It keeps effective challenge, performed by people with enough independence to stay objective. Generative and agentic AI fall outside it, so a bank should not read compliance with SR 26-2 as covering them. In Kriv AI's practice, put them in the model inventory under the bank's own governance, with an owner and a risk tier for each, and report on them to the board.

    SR 11-7 compliance: validation and vendor models under SR 26-2

    For SR 11-7 compliance questions about validation, the current text is SR 26-2. It says validation evaluates whether a model performs as expected, with rigor that follows the model's approach, use and materiality, and it names three components: conceptual soundness, outcomes analysis and ongoing monitoring. Validation generally occurs before a model's first use. If an urgent business need means a model is used earlier, SR 26-2 points to closer attention to its limitations and to controls such as limits on use or closer monitoring. For vendor products it says the principles still apply even when the vendor does not share code, data or methodology, and it treats validation of vendor products, by internal or outside parties, as an important element.

    roadmap

    A Compliance Roadmap, in Sequence

    The sequence: first, inventory every AI system currently in production or pilot, including vendor tools with embedded AI. Next, risk-tier each one and identify validation gaps against the bank's own model risk standards. Then close the highest-risk gaps first and stand up the ongoing governance structure to keep the inventory current.

    engagement

    How Kriv AI Implements This

    For illustration, not a measured client result: a regulated financial services client engaged Kriv AI to build an AI model inventory and validation framework ahead of an SR 26-2 gap assessment, the kind of anonymized engagement pattern we bring to a scoped model risk management project: inventory, risk tiering, validation framework design, and an ongoing governance structure your board can actually report against.

    Straight answers

    Frequently asked questions about Model Risk Management for AI: From SR 11-7 to SR 26-2

    What is SR 26-2?

    SR 26-2 is the Federal Reserve, OCC, and FDIC model risk management guidance that superseded SR 11-7 in April 2026. It excludes generative AI and agentic AI models from its scope, describing them as novel and rapidly evolving, and it does not set forth enforceable standards. SR 26-2 footnote 1 says: "See 12 CFR Part 4, Subpart F, Appendix A (OCC); 12 CFR Part 262, Appendix A (Board); 12 CFR Part 302, Appendix A (FDIC). However, supervisory action may result for any violations of law or unsafe or unsound practices stemming from insufficient management of model risk."

    Does SR 11-7 still apply, or has it been replaced?

    SR 11-7 was superseded by SR 26-2 in April 2026. SR 26-2 does not extend to generative and agentic AI, so a bank's own risk management and governance practices should guide the controls for those systems (footnote 3).

    Does SR 11-7 apply to large language models?

    The original SR 11-7 guidance predates LLMs and does not address them explicitly, and SR 26-2 does not close that gap: it excludes generative and agentic AI models from its scope. A bank's own risk management and governance practices should guide the controls for those systems (footnote 3).

    How do you validate an LLM if you can't backtest it like a traditional model?

    Through structured output sampling against a rubric, red-teaming for guardrail failures, and ongoing drift monitoring on output quality, rather than a single input-output accuracy metric.

    What should an AI model inventory include?

    Every AI system in production or pilot, including third-party vendor tools with an embedded LLM, each assigned a risk tier based on the decision it influences. For generative and agentic AI this comes from the bank's own governance, not from SR 26-2.

    What does an SR 26-2 gap assessment involve?

    The sequence: inventory the AI systems first, risk-tier them and identify gaps next, then close the highest-risk gaps and stand up ongoing governance.

    Does SR 26-2 apply to generative AI?

    No. SR 26-2 says generative AI and agentic AI models are novel and rapidly evolving, so they are not within the scope of the guidance (footnote 3). The same footnote says a banking organization's own risk management and governance practices should guide the controls for tools the guidance does not cover. The guidance does not set forth enforceable standards, and it says non-compliance with it will not result in supervisory criticism. Footnote 1 adds that supervisory action may result from violations of law or unsafe or unsound practices stemming from insufficient management of model risk. A bank should therefore still decide its own governance and controls for large language models and AI agents. Kriv AI's approach, informed by SR 26-2, starts with an inventory entry, an owner and a risk tier for each one.

    Is SR 11-7 compliance still required?

    SR 11-7 is no longer the current guidance. The Federal Reserve's letter of April 17, 2026 says SR 26-2 supersedes and replaces SR 11-7, which was issued on April 4, 2011. SR 26-2 does not set forth enforceable standards or prescriptive requirements, so it is guidance, not a rule. The Federal Reserve letter says it is expected to be most relevant to banking organizations with over $30 billion in total assets regulated by the Federal Reserve, and the OCC bulletin says it also applies to community banks, subject to the limits in the guidance. It takes a risk-based approach and covers effective challenge by people with enough independence, a model inventory, and validation whose rigor follows the model's approach, use and materiality. Which controls your bank applies is a decision for its own model risk team and its supervisors.

    Does SR 26-2 apply to machine learning models?

    Footnote 3 of SR 26-2 says its principles apply to traditional statistical and quantitative models and to non-generative, non-agentic AI models. The guidance defines a model as a complex quantitative method, system or approach that applies statistical, economic or financial theories to process input data into quantitative estimates. It excludes simple arithmetic calculations, such as those in spreadsheets, and deterministic rule-based processes. Whether a particular machine learning model fits that definition is a judgment for the bank's model risk function, so we do not rule individual models in or out here. Generative and agentic AI are outside the guidance. Our inventory approach, informed by SR 26-2, lists every AI system, including vendor tools, so the bank can record which side of that line each one sits on.

    Does SR 11-7 apply to machine learning models?

    SR 11-7 no longer applies to machine learning models or any other models, because SR 26-2 superseded it on April 17, 2026. Footnote 3 of SR 26-2 says its principles apply to traditional statistical and quantitative models and to non-generative, non-agentic AI models. Whether a particular machine learning model fits the guidance's definition of a model is covered in the answer to the question above.

    What was SR 11-7?

    SR 11-7 was the Federal Reserve supervisory letter titled Guidance on Model Risk Management, issued on April 4, 2011. SR 26-2, issued on April 17, 2026 by the Federal Reserve, the OCC and the FDIC, supersedes and replaces it, together with SR 21-8, the 2021 interagency statement on model risk management for systems supporting Bank Secrecy Act and anti-money laundering compliance. We describe SR 11-7 only as the SR 26-2 letter describes it. If a policy or contract of yours still cites SR 11-7, check it against SR 26-2. Model risk management for generative and agentic AI is covered in the sections above.

    What is effective challenge under SR 26-2?

    SR 26-2 describes effective challenge as critical analysis by objective experts who evaluate model risk and push for appropriate changes across the model lifecycle, from development to ongoing monitoring. It says the people doing it need the expertise for critical and objective challenge, enough independence to stay objective, and the organizational standing and influence to effect change. Who fills that role in your bank is for your own model risk function to decide.

    What is model materiality in SR 26-2?

    Materiality is how much a model matters to the bank, and SR 26-2 builds it from two things: exposure, the significance of the model's output to business decisions, and purpose, the nature and importance of the model's use. Models judged more material warrant more comprehensive and rigorous oversight. For models a bank deems immaterial, SR 26-2 says risk management may consist of identifying them and monitoring whether conditions make them material later.

    Does SR 26-2 cover vendor models?

    Yes, as third-party products. SR 26-2 has a section on vendor and other third-party products, including data, parameter values or complete models. It says the principles of model risk management remain applicable even when the vendor does not provide code, data or methodology, and that sound practice includes understanding the vendor model and monitoring its performance. That is separate from the scope footnote, which leaves generative and agentic AI out of the guidance.

    How often does SR 26-2 say a model must be validated?

    It sets no fixed schedule. SR 26-2 says the timing, nature and frequency of validation vary with model purpose, methodology, how often and how far the model changes, data limits and other practical constraints, and that validation generally occurs before first use. The OCC bulletin adds a note for community banks, quoting OCC Bulletin 2025-26, that OCC guidance on model risk management does not require community banks to perform annual model validation.

    Does SR 26-2 require a model inventory?

    SR 26-2 does not use the word require here. It describes a comprehensive set of information on models under development or in use as common industry practice, and says an effective inventory has enough information to understand model risk at the individual and aggregate levels. The detail can vary with model complexity and how much a bank uses models. For generative and agentic AI, which SR 26-2 leaves out of scope, an inventory entry comes from the bank's own governance.

    What does internal audit do under SR 26-2?

    SR 26-2 says that where internal audit is part of a bank's model risk management practices, it would generally not duplicate activities such as model development or validation. Internal audit's role is generally to evaluate whether the model risk management practices are rigorous and effective and whether related policies are implemented accordingly. The guidance also says that when a bank uses external resources to help manage model risk, sound practice involves maintaining proper oversight and integrating that work into broader model risk management. How your bank splits these roles is for its own model risk team and its supervisors.

    Can an AI platform handle SR 26-2 compliance?

    SR 26-2 says it does not set forth enforceable standards or prescriptive requirements, and it says each banking organization is responsible for adopting model risk management practices that suit its own risks. It describes a model inventory and documentation as supporting effective model risk management, so a tool that holds those can help with them. No tool makes a bank meet SR 26-2. SR 26-2 also puts generative and agentic AI models outside its scope, so check what any vendor says its product covers. We do not rank vendors here.

    Talk to the team that would do the work

    Bring your requirements to a working session with the person who'll actually deliver.

    Book a Discovery Call