Agentic AI in banking is moving into production faster than the supervisory guidance that governs it. The revised interagency model risk management guidance issued on April 17, 2026 states that generative AI and agentic AI models are not within its scope. The October 2026 Evident AI Index shows how leading banks respond: 80% of the top-10 banks show public evidence of controls for specific AI use cases, against 40% of the other 40 banks (Evident AI Index: Key Findings Report, October 2026). Evident reports that advances in agentic AI are driving investment in automated guardrails and post-deployment monitoring. For a mid-size bank, the question is which controls to build and in what order. This article covers what changed in 2026, what leading banks do, and a control path a mid-size bank can start now.
What Changed for Agentic AI in Banking in 2026
What an AI Agent Changes for Governance
A model produces an estimate that a person or a process then uses. An agent goes further: it takes actions through the tools and permissions it holds, so it can query systems, change records, send messages, or start transactions, depending on how it is configured. That changes what a bank must control. Accuracy of the output still matters, but so do what the agent is allowed to do, what data it can reach, and when a person steps in. A bank that governs agents only through its model inventory misses these actions entirely. Evident describes the same shift, with governance moving from validating a model toward controlling the whole workflow around it.
The April 2026 Model Risk Guidance Leaves Agentic AI Out
On April 17, 2026, the Federal Reserve, the OCC, and the FDIC jointly issued revised model risk management guidance (Federal Reserve SR 26-2 and OCC Bulletin 2026-13). It replaces SR 11-7 and the 2021 interagency statement on model risk for BSA and AML systems. A footnote states that generative AI and agentic AI models are novel and rapidly evolving, so they are not within the scope of the guidance. The principles still apply to traditional statistical models and to non-generative, non-agentic AI models, which means a machine learning credit model stays in scope while an AI agent does not. Elevate’s analysis of the SR 11-7 rescission covers the guidance in full.
The Threshold and the Footnote Matter Together
The guidance is expected to be most relevant to banking organizations with more than $30 billion in total assets. For banks at or below that size, models typically fall under internal risk management and governance practices suited to the bank’s size and risk profile, although the guidance may be relevant where model exposure is significant. It also sets no enforceable standards, and the agencies state that non-compliance will not result in supervisory criticism. The footnote adds that a bank’s own risk management and governance practices should guide the controls for tools the guidance does not cover. For a mid-size bank, the exclusion therefore does not remove the expectation of governance. It moves the work to the bank.
The February 2026 FS AI RMF Offers a Control Set
On February 19, 2026, the US Treasury released the Financial Services AI Risk Management Framework (FS AI RMF) alongside an AI lexicon. The Cyber Risk Institute hosts the framework and its 230 control objectives, which adapt the NIST AI RMF to financial services. It opens with an adoption stage questionnaire that places an institution in one of four stages: Initial, Minimal, Evolving, or Embedded. The stage sets the scope, from 21 control objectives at Initial to 126 at Minimal, 193 at Evolving, and all 230 at Embedded. The framework is voluntary, and Elevate’s FS AI RMF readiness service helps institutions determine their stage and scope the controls that follow.
What Leading Banks Do About Agentic AI Controls
Principles Are Universal, Evidence of Controls Is Not
The Evident AI Index benchmarks 50 of the largest banks in North America, Europe, and Asia using only publicly available data. Responsible AI principles are now almost universal: 49 of the 50 banks publish them, up from 16 in 2023. A majority (62%) explain how those principles move into practice, but fewer than half (48%) share evidence of controls implemented for a specific use case. That group skews heavily toward the top-10 banks: 80% of them show such evidence, against 40% of the other 40. Because the Index counts only what banks publish, the gap measures public evidence, not the controls themselves.
Governance Moves From the Model to the Workflow
Evident observes that governance changes when systems take actions instead of only generating answers. The focus moves beyond conventional model risk, meaning bias, explainability, and validation, toward controlling the whole workflow. The controls it lists are grounding, input and output validation, tool permissions, human intervention, and continuous monitoring. Evident also describes leading banks as investing in controls as an enabler of scale, not only as a brake on risk. Banks that build controls ahead of deployment can release more autonomous systems faster, without governance becoming the next bottleneck.
Controls Are Built Once and Reused
Leading banks treat the controls as infrastructure around the model. Evident describes this as a reusable “harness” of proprietary data, integrations, tools, permissions, evaluations, controls, and orchestration, so that each deployment makes the next one faster, cheaper, and easier to govern. The same logic supports routing tasks to different models by complexity, latency, risk, and cost. For a mid-size bank, the lesson is structural: a control set written once and applied to every agent costs less than one written for each pilot. Reuse also helps with evidence: a shared logging and approval layer produces the same records for every agent, which makes testing simpler.
Model Risk Talent Grows With Deployment
Staffing follows the same pattern. AI model risk roles grew 9.0% year over year across 127,000 AI and data employees, behind only enablement (21.2%) and product management (10.0%). Evident links that growth to the need to manage increasingly complex risks as deployment scales (Evident AI Index: Key Findings Report, October 2026).
What the Index Does Not Show
The Index is an outside-in benchmark of public evidence, and it has limits. It covers 50 of the largest banks, so its results describe large institutions, and a community or regional bank should read them as direction, not as a peer comparison. It counts what a bank publishes, so a bank with strong controls and little disclosure scores lower than its practice deserves. It also shows that leaders publish more evidence, not that publishing causes better controls. Read the 80% and 40% figures as a signal of where public disclosure is heading, and use them to frame the questions a bank should be able to answer internally, not as a measure of safety.
Agentic AI Controls a Mid-Size Bank Can Build
Start With an AI Inventory
Control starts with knowing what runs. Because agentic AI models sit outside the model risk guidance, they may never reach a bank’s model inventory, and agents and copilots often arrive inside vendor releases rather than through an AI project. An AI inventory should list each agent, copilot, and AI feature in a vendor product, with its owner, the data it reaches, the tools and permissions it holds, and whether the bank built it or a vendor supplied it. That record is the base for every later decision, including the adoption stage. The model risk guidance sets a useful standard here: it describes an effective model inventory as holding enough information to understand model risk for each model and in aggregate, and an AI inventory should meet the same test. Elevate’s guide to AI governance for banking compliance covers the program around it.
Ask Vendors What Their Embedded Agents Do
Many banks will meet agentic AI first through a vendor product. Before an agent runs inside a procured system, the bank should ask what actions it can take, which data it reads and writes, where a person approves its work, and what logs the bank can see. It should also ask how the vendor notifies customers of changes to the underlying models and how it reports an AI incident. The answers belong in the contract and in the inventory, because they decide what the bank can verify later. The FS AI RMF gives substantial attention to third-party AI, including due diligence on data provenance and the contractual provisions that follow from it.
Determine the Adoption Stage Before Scoping
The FS AI RMF questionnaire runs from Embedded downward, and a single matching statement ends the assessment at that stage. Evolving describes AI that drives outcomes through external-facing solutions and sensitive data without yet making critical decisions on its own. A bank whose agent touches customer data or customer-facing outcomes should therefore expect to land at Evolving, which scopes 193 control objectives, and not at Minimal, which scopes 126. Building to the higher stage and phasing the work is easier to defend than writing to the lower stage and revising later. The stage is not a label to negotiate. It decides how much of the control set applies.
Define Oversight, Monitoring, and Shutdown for Each Agent
Leading banks control the workflow, not only the model. For each agent, a bank should document the actions it may take, the tools and data it may reach, the point where a person must approve, the thresholds that trigger a stop, and who declares an AI incident and how it joins the existing incident process. Those items correspond to three of the controls Evident lists: tool permissions, human intervention, and continuous monitoring. Grounding and input and output validation sit in the technical build of the agent. Elevate’s guide to governing autonomous AI agents covers the threat side.
Sequence the Work in Three Passes
The work runs best in three passes. First, build the AI inventory and determine the adoption stage, because every later scope decision depends on them. Second, write controls for the agents with the most access or the most customer impact, and phase the rest by stage. Third, produce the monitoring evidence and run a first internal audit test of the controls, so gaps surface before an examiner finds them. Each pass should end with a written decision that names an owner and a date, so the program does not stall between steps. Elevate recommends re-running the adoption stage questionnaire every year, because internal model development is the trigger that moves an institution to Embedded.
Make Every Control Testable
A control set earns trust when internal audit can test it. Each control statement should carry its domain, a named owner, a frequency, the evidence it produces, and its anchor in the framework. A statement that does not name its evidence cannot be tested, and one without an anchor cannot be defended to an examiner who asks why the requirement exists. The AI governance intake questionnaire gives a bank a starting register: 136 questions across ten stakeholder groups and thirteen control domains, each anchored to a CRI FS AI RMF control objective. Download the AI governance intake questionnaire for banks to see where coverage stands before an examiner or a counterparty asks.
Conclusion
Agentic AI in banking sits in a gap the 2026 guidance left open. The April 2026 model risk guidance excludes generative and agentic AI, the February 2026 FS AI RMF supplies a voluntary control set, and the Evident AI Index shows leading banks already building the controls: 80% of the top-10 banks show public evidence of them, against 40% of the rest.
For a mid-size bank, the practical path is short: inventory the agents and AI features, determine the adoption stage, document oversight and shutdown for each agent, and make each control testable. Elevate Consult prepares institutions for examinations and audits performed by others and does not examine or certify them. Book a call with an Elevate advisor to place your institution in the framework and plan the build.
Key Takeaways
The 2026 guidance leaves agentic AI to each bank’s own governance, so the controls are a bank decision.
- The April 2026 guidance excludes agentic AI. SR 26-2 and OCC Bulletin 2026-13 place generative and agentic AI models outside their scope, and the guidance is expected to be most relevant to banks above $30 billion in assets.
- The FS AI RMF supplies a voluntary control set. Its 230 control objectives scale by adoption stage, from 21 at Initial to 193 at Evolving.
- Leaders show evidence, not only principles. 80% of the top-10 banks show public evidence of use-case controls, against 40% of the other 40 (Evident AI Index, October 2026).
- Controls move from the model to the workflow. Tool permissions, human intervention, and continuous monitoring apply to each agent.
- Testable controls hold up. Each control statement needs an owner, a frequency, and the evidence it produces.
Frequently Asked Questions
Does the April 2026 model risk guidance cover agentic AI?
No. The revised guidance issued on April 17, 2026 (Federal Reserve SR 26-2 and OCC Bulletin 2026-13) states that generative AI and agentic AI models are novel and rapidly evolving, so they are not within its scope. It says a bank’s own risk management and governance practices should guide the controls for tools it does not cover. Its principles still apply to traditional statistical models and to non-generative, non-agentic AI models.
Does the model risk guidance apply to a mid-size bank?
The guidance is expected to be most relevant to banking organizations with more than $30 billion in total assets. Models at banks with $30 billion or less typically fall under internal risk management and governance practices suited to the bank’s size and risk profile, although the guidance may be relevant where model exposure is significant. The guidance is not enforceable, and the agencies state that non-compliance will not result in supervisory criticism.
What is the FS AI RMF and how many controls does it include?
The Financial Services AI Risk Management Framework was released by the US Treasury on February 19, 2026, and the Cyber Risk Institute hosts its 230 control objectives. It adapts the NIST AI RMF to financial services and begins with an adoption stage questionnaire. The stage sets the scope: 21 control objectives at Initial, 126 at Minimal, 193 at Evolving, and all 230 at Embedded. The framework is voluntary.
What did the Evident AI Index find about AI controls in banks?
The October 2026 Evident AI Index, which benchmarks 50 of the largest banks using public data, found that 49 publish responsible AI principles and 48% share evidence of controls for specific use cases. The share is 80% among the top-10 banks and 40% among the other 40. Evident reports that advances in agentic AI are driving investment in automated guardrails and post-deployment monitoring. The Index measures public evidence, not the controls themselves.
Where should a mid-size bank start with agentic AI controls?
Start with an AI inventory of every agent, copilot, and AI feature in a vendor product, with its owner, data, tools, and permissions. Then determine the adoption stage with the FS AI RMF questionnaire and document oversight, monitoring, and shutdown for each agent. Write each control with an owner, a frequency, and the evidence it produces so internal audit can test it.