Model inventory
A record of every model in use, with enough information to understand its risks. Materiality, set by a model’s purpose and exposure, decides how much scrutiny it gets.
AI inventoryHow banks govern the models behind their decisions, what US regulators changed in April 2026, and where generative and agentic AI now sit.
Model risk management is the set of governance, development, validation and monitoring practices an organization uses to control the harm that wrong or misused models can cause.
It grew out of banking supervision and covers every model that informs a decision, built in-house or bought from a vendor, for as long as that model is in use.
in total assets: the size above which the 2026 US guidance is expected to be most relevant
Source: Federal Reserve, SR 26‑2
principles in the PRA’s SS1/23, from model identification to model risk mitigants
Source: Bank of England, PRA SS1/23
Two texts anchor most programs today, and their wording decides what lands in your inventory.
| US revised guidance (SR 26‑2) | UK PRA SS1/23 | |
|---|---|---|
| Issued by | Federal Reserve, OCC and FDIC | Prudential Regulation Authority |
| In effect | April 17, 2026, replacing SR 11‑7 | May 17, 2024; current version April 23, 2026 |
| A model is | A complex quantitative method, system or approach that applies statistical, economic or financial theories to produce quantitative estimates | A quantitative method, system or approach that applies statistical, economic, financial or mathematical theories, techniques and assumptions |
| On AI | Non-generative, non-agentic AI models are covered. Generative and agentic AI models are not | Addresses AI techniques such as machine learning, as far as they apply to models generally |
| Written for | Banking organizations, most relevant above $30 billion in total assets | UK banks, building societies and designated investment firms with internal model approval |
The US guidance covers development and use, validation and monitoring, governance and controls, and vendor products. In practice that means four working parts.
A record of every model in use, with enough information to understand its risks. Materiality, set by a model’s purpose and exposure, decides how much scrutiny it gets.
AI inventoryDocumented design, data and assumptions, plus users who understand a model’s limitations and monitor its performance.
An evaluation of whether each model performs as expected, generally before first use and again as the model changes.
Model validationCritical analysis by objective experts with the standing to bring about change. SR 11‑7 called it a guiding principle of model risk management.
This is the question model risk teams are working through in 2026.
A footnote in the revised guidance calls generative and agentic AI models novel and rapidly evolving, and not within its scope. A bank’s own governance should set controls for what it does not cover.
Read SR 26‑2The agencies plan a request for information that will consider banks’ use of AI, including generative and agentic AI.
OCC Bulletin 2026‑13Proprietary code is no exemption. The guidance expects vendor products to be validated, by internal or outside parties, and monitored for fitness for purpose.
Cranium AI is not a validation workbench and ships no banking guidance as a built-in framework. Its AI‑BOMs, AI Cards, test results and runtime records support a bank’s own model risk process.
Banking model risk managementStraight answers, kept current. Last updated .
Ask us anything →Yes. On April 17, 2026 the Federal Reserve, OCC and FDIC issued revised guidance on model risk management. The Federal Reserve’s letter, SR 26‑2, supersedes SR 11‑7 from April 2011, and OCC Bulletin 2026‑13 rescinds OCC Bulletin 2011‑12. The OCC says the new guidance does not set enforceable standards or prescriptive requirements.
Under the US guidance, not directly. It states that generative and agentic AI models are not within its scope, while non-generative, non-agentic AI models are. The risk remains: a bank’s own risk management and governance practices should set controls for tools the guidance does not cover.
The formal guidance is written for banks. The US text is most relevant to banking organizations with more than $30 billion in total assets. The practices travel well: any organization that relies on models can keep an inventory, validate independently and monitor. For AI more broadly, see AI risk management.
Primary sources for the figures and frameworks cited on this page. Last verified .
See how Cranium AI discovers, observes, governs, secures and proves every AI model, agent and vendor in your enterprise.
Cookie Policy · Privacy Policy
Global Privacy Control detected. Opt-out request honored: marketing tags stay off in this browser.