Skip to main content
    Kriv AI

    Bank AI Model Risk

    Model validation: what it is, what it covers and who does it

    A plain guide to model validation in a bank. It explains what a validation checks, when it happens, who can carry it out, and where AI models fit under the current guidance.

    Model validation is the part of model risk management that checks whether a model performs as expected and assesses how reliable it is and where it falls short. Under SR 26-2, the revised guidance issued in April 2026 by the Federal Reserve, the OCC and the FDIC, validation looks at conceptual soundness, outcomes analysis and ongoing monitoring. It generally happens before a model's first use, and its nature and rigor generally align with the model's approach, use and materiality. The guidance says the quality of validation depends on the rigor and effectiveness of the review rather than on the organizational structure of the bank's risk management function.

    context

    What Is Model Validation?

    SR 26-2, titled Revised Guidance on Model Risk Management and issued on April 17, 2026, says model validation evaluates whether models perform as expected and includes an assessment of a model's reliability and its limitations. Sound validation identifies model limitations and errors, clarifies appropriate use, and shows whether corrective actions may be warranted. SR 26-2 supersedes and replaces SR 11-7, so anyone who learned the topic under the older letter should read the newer one.

    Validation is not a single test. It draws on a model's assumptions, methods, data and relevant theories to describe where the model risk comes from and how large it is. It can also reveal performance deterioration over time and inform judgments about what range of performance is acceptable. When performance departs meaningfully from expectations, the guidance says banks generally consider whether the model needs adjustment, recalibration or redevelopment.

    tiers

    What a Model Validation Covers

    SR 26-2 describes three components of model validation. They are listed here in the order the guidance presents them.

    1. 1. Conceptual soundness

      Assessing and documenting the model's design, including key modeling choices, assumptions, qualitative judgments and data selection, plus its construction and developmental testing. The guidance notes that for some models, interpretability measures or benchmarking against other models may be more practical than evaluating theoretical construction.

    2. 2. Outcomes analysis

      Comparing model outputs with real world outcomes to see how the model performs against its objectives and business use. It can take the form of testing during development, reports from ongoing monitoring, or standalone work such as back testing or outlier analysis.

    3. 3. Ongoing model monitoring

      Evaluating whether the model keeps performing as expected as products, exposures, activities, clients, data relevance or market conditions change. How often monitoring reports are produced, and what they cover, depends on the nature of the model, the availability of new data or modeling approaches, and the model's materiality.

    context

    When Validation Happens

    The guidance says validation generally occurs before a model's first use. It also recognizes that an urgent business need may mean a model is used before validation is finished. In that case it describes sound practice as paying greater attention to the model's limitations, telling the relevant stakeholders about them, and setting controls such as limits on use or closer monitoring of performance.

    After first use, the timing, nature and frequency of validation work vary with the model's purpose and methodology, how often and how widely it changes, data limitations and other practical constraints. The guidance adds that even with rigorous validation, material model risk can remain, so people who use model output benefit from understanding its limitations and supplementing it with other analysis.

    capabilities

    Who Validates a Model

    SR 26-2 is more concerned with how good the review is than with who employs the reviewer.

    Rigor and expertise matter more than structure

    The guidance says the quality of the validation process depends on the rigor and effectiveness of the review rather than on the organizational structure of the bank's risk management function. Technical knowledge and modeling expertise are what allow meaningful critique and sound judgment about model performance and risk.

    Watch for conflicts of interest

    The guidance says model risk management benefits from clear roles and well defined accountability, including around potential conflicts of interest. Its example is misaligned incentives between reporting lines, such as model development and validation groups.

    Internal audit has a different job

    Where internal audit is part of the program, the guidance says it would generally not duplicate activities such as model development or validation. Its role is generally to evaluate whether the model risk management practices are rigorous and effective and whether related policies are implemented accordingly.

    Outside parties and vendor models

    For banks that use external resources to help manage model risk, the guidance describes sound practice as keeping proper oversight and folding that work into the wider program. It also calls the validation of vendor products an important element of model risk management, which may be done by internal or outside parties, even though a vendor may not hand over its underlying code, data or methodology.

    ai gap

    Which Models and AI Systems the Guidance Covers

    SR 26-2 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 and software with no such theories underneath.

    On AI, the guidance says generative AI and agentic AI models are novel and rapidly evolving and are not within its scope, while saying a bank's own risk management and governance practices should guide controls for tools and systems it does not cover. It states that its principles apply to traditional statistical and quantitative models and to AI models that are neither generative nor agentic. The guidance leaves the approach for those tools to the bank's own risk management and governance practices.

    Two more points affect who needs to read it. The guidance is expected to be most relevant to the largest banking organizations, above an asset threshold the Federal Reserve sets (our SR 26-2 page gives the figure), though it may also be relevant to smaller banks with significant exposure to model risk. It also says it does not set enforceable standards or prescriptive requirements, while noting that supervisory action may still follow violations of law or unsafe or unsound practices that stem from insufficient management of model risk. Ask your counsel how that applies to your bank.

    differentiation

    How This Differs From Our Related Pages

    This page answers the general question of what model validation is, what it covers and who does it. For how a bank scales the depth of review to risk, see our page on how banks tier validation by materiality. For the move from the old letter to the new one, see our page on model risk management under SR 26-2 and our page on SR 11-7, the letter that SR 26-2 superseded. For the wider discipline, see our page on what AI model risk management means for banks, and for why AI models struggle in review, see our page on why bank AI models fail model risk review. Use this page when the question is what validation itself involves.

    engagement

    How an Engagement Works

    We start with a scoping call about which models your bank runs, which of them use machine learning or generative AI, and who owns each one. We then agree a defined scope and deliverables, work alongside your model risk, compliance, and technology leaders, and hand over artifacts you own. We do not replace your independent validation function, give legal advice, or tell you how a regulator will read your program. Every engagement is scoped and priced after a short discovery call.

    get a quote

    How to Get a Real Quote

    Book a discovery call and we will scope the work with you.

    Straight answers

    Frequently asked questions about Model validation: what it is, what it covers and who does it

    What is model validation?

    SR 26-2 says model validation evaluates whether models perform as expected and includes an assessment of a model's reliability and its limitations.

    What does a model validation cover?

    SR 26-2 describes three components: conceptual soundness, outcomes analysis and ongoing model monitoring.

    When should a model be validated?

    The guidance says validation generally occurs before a model's first use. If a model must be used earlier, it describes sound practice as paying greater attention to its limitations, informing stakeholders and setting controls such as limits on use or closer monitoring.

    Who can validate a model?

    The guidance says the quality of validation depends on the rigor and effectiveness of the review rather than on the organizational structure of the bank's risk management function. It also says validation of vendor products can be done by internal or outside parties.

    Does SR 26-2 cover generative AI models?

    No. The guidance says generative AI and agentic AI models are not within its scope, and that its principles apply to traditional statistical and quantitative models and to AI models that are neither generative nor agentic.

    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