Information Security & Compliance Blog - Tandem

About the OCC’s New Super Intelligence: Supervisory Framework

Written by Savannah Richardson | Nov 18, 2026, 5:56:52 PM

On November 4, 2026, the Office of the Comptroller of the Currency (OCC) released its Super Intelligence: Supervisory Framework. The framework is intended to help bank examiners evaluate how financial institutions manage risks associated with artificial intelligence (AI), particularly generative and agentic AI. 

 Note: You may have noticed a new term appearing in federal government publications: Super Intelligence (SI). Following Executive Order 14434, federal agencies have begun using SI in place of AI. While the terminology has changed, the technology we're discussing hasn't. For familiarity, we'll continue using AI throughout this article. 

The OCC’s new framework offers a look at how examiners may evaluate AI-related risks and the controls financial institutions have in place to manage them. Rather than introducing an entirely new set of requirements, the framework provides supervisory considerations that build on existing risk management practices.

So, what does the framework cover, and what should financial institutions take away from it? Let's take a closer look.

In This Article

How Is the Framework Organized?

The OCC's framework is organized into six pillars and 16 components. Each pillar addresses an area of AI risk management, with supporting key concepts and examples of supervisory considerations.

 Pillar   Principal Supervisory Consideration (Simplified) 

Purpose

What the AI system is intended and permitted to do.

Risk

What could go wrong and the potential consequences.

Control

What safeguards are in place to manage AI-related risks.

Oversight

Who is responsible for overseeing AI systems and activities.

Resilience

How the institution prepares for and responds to AI-related disruptions.

Assurance

How the institution evaluates whether AI systems and controls are working as intended.

Unlike some cybersecurity and risk management frameworks, the OCC's framework does not provide a standardized set of assessment statements, maturity levels, or scoring criteria. Instead, it outlines considerations that examiners may use when evaluating an institution's AI risk management practices. In simple terms, this guidance is more like a glossary than a recipe.

For example, the framework discusses defining an AI system’s purpose and permitted use (key concept). This includes documenting what the system is supposed to do, what information it can access, what actions it can take, and when humans need to apply oversight (examples). The OCC also suggests capturing some of these details in an AI inventory.

These examples give institutions an idea of what examiners may look for, but they aren’t a prescribed list of controls. Institutions will still need to determine how the considerations apply to their AI use cases and what documentation or controls make sense for their level of risk.

The OCC also describes the framework as a living document, allowing it to evolve as AI technology, use cases, and associated risks change.

One interesting detail worth mentioning is that the OCC acknowledged using AI to help prepare the framework itself. The document includes a disclaimer stating that the content was reviewed, edited, and approved by a human. A fitting example of human oversight in a framework that spends quite a bit of time discussing it!

Key Takeaways from the OCC Super Intelligence: Supervisory Framework

Build on Existing Risk Management Practices

One of the OCC's central messages is that financial institutions do not necessarily need to create an entirely new risk management program for AI. Instead, institutions can incorporate AI-related risks into their existing governance, risk management, and compliance practices.

The OCC also clarifies that institutions are not required to establish a separate AI risk category or risk appetite. However, the framework does include considerations for defining an institution's tolerance for AI-related risks, particularly those involving autonomous actions, customer impact, and operational exposure.

For example, an institution may already have established risk tolerances for financial losses, operational disruptions, or unauthorized transactions. Those existing considerations can help determine whether an AI use case introduces additional risks or requires new safeguards.

The key is understanding how AI changes an institution's existing risk exposure rather than assuming every AI-related risk requires an entirely new process. 

Pay Attention to Agentic AI

While the framework addresses AI broadly, much of its focus is on generative and agentic AI.

Generative AI can create text, images, and other content. Agentic AI takes this a step further by using AI systems that can plan, make decisions, and perform actions with varying levels of autonomy.

For financial institutions, this distinction matters because the risks can change when an AI system moves from providing information to taking action.

Consider an AI tool that summarizes customer service interactions. The primary concerns may involve the accuracy of the summary, data privacy, and appropriate use of customer information.

Now consider an AI agent that can independently update customer records, initiate transactions, or interact with other systems. In addition to accuracy and privacy, the institution must consider what actions the agent is authorized to perform, how those actions are monitored, and when human intervention is necessary.

This helps explain why the OCC places considerable attention on permitted actions, system permissions, human oversight, and the ability to suspend or override AI activities.

Establish Meaningful Human Oversight

Human oversight is another recurring theme throughout the framework. However, oversight involves more than simply having a person review an AI system's outputs or actions.

The OCC highlights several considerations, including whether employees understand the AI systems they oversee, have the authority to intervene, and know when an issue should be escalated.

The answers will depend on the use case. Reviewing an AI-drafted email before it goes out is very different from overseeing an agent that can perform underwriting tasks. The latter may call for transaction limits, approval requirements, monitoring, and clear escalation procedures. The OCC also discusses documenting overrides and increasing human involvement when a system behaves in an unexpected way.

Not every AI action needs the same approval process. What matters is whether the level of oversight makes sense for the potential impact.

Account for AI in Third-Party Relationships

An institution doesn’t have to build an AI system to use one. AI features may already be part of products it gets from vendors, and those features can change over time.

Say a fraud detection vendor adds an AI feature to its platform that analyzes transactions. The institution may need to understand how the feature works, what information it uses, and whether changes to its detection methods could affect how suspicious transactions are identified. Existing due diligence, contract review, and ongoing monitoring processes are a good place to start.

It’s also worth understanding whether the vendor depends on another AI or cloud provider. A disruption farther down the supply chain can still affect the institution’s operations.

Prepare for AI-Related Disruptions

AI-related disruptions may look different from the technology failures financial institutions usually plan for. While an AI system could become unavailable due to a service outage, it could also continue operating while producing inaccurate information or taking unintended actions.

The OCC's framework emphasizes operational resilience, including the ability to identify problems, intervene when necessary, and maintain or restore critical business operations.

For example, a financial institution may rely on an AI-powered customer service platform to assist with routine customer inquiries. If the platform becomes unavailable or begins providing incorrect information, employees may need to take over those responsibilities or use alternate procedures to continue serving customers.

Institutions should consider these types of dependencies when reviewing their business continuity and disaster recovery plans. This may include identifying fallback procedures, establishing conditions for suspending AI functionality, and testing whether critical business processes can continue without the AI system.

The goal is not only to restore the technology, but also to make sure the institution can continue operating safely when AI systems experience issues.

Apply a Risk-Based Approach to AI Controls

The OCC also includes an illustrative tool that organizes AI systems and agents into four risk tiers. These tiers consider factors such as an AI system's autonomy, the actions it can perform, and the potential consequences of those actions.

Tier

Profile Characteristics

Use Case

Tier 1: Critical

Direct action, low/no reversibility, high financial or customer impact, autonomous or partially autonomous

Autonomous payment agent; real-time credit decisioning; fraud-hold release; market access

Tier 2: High

Direct action or high-weight decision support, human-on-the-loop, moderate-to-high impact

SI-assisted underwriting; agent-initiated customer communications; AML alert disposition; code assistance for internal code or data access changes

Tier 3: Moderate

Decision support with mandatory human-in-the-loop, limited data sensitivity, reversible

Internal document summarization for credit memos; agent-assisted research

Tier 4: Low

No customer/financial data exposure, no direct action, fully human-controlled use

Internal productivity tools; code assistance; general employee Q&A

3.2 Illustrative Risk Tier Definitions from the OCC Supervisory Framework on Bank Use of Super Intelligence.

The framework does not establish a mandatory scoring methodology or require institutions to adopt these specific tiers. Instead, it provides an example of how institutions might evaluate AI use cases and apply controls based on their associated risks.

Not every AI system requires the same level of oversight, but institutions should be able to explain how they determine which safeguards are appropriate.

 

How Financial Institutions Can Prepare

While the OCC's framework is intended for examiners, financial institutions can use its supervisory considerations to review their existing AI risk management practices and identify opportunities for improvement.

Here are a few areas to consider:

    • Maintain an AI inventory. Identify where AI is being used throughout the institution, including employee tools, internal systems, customer-facing applications, and AI capabilities embedded in third-party products. Document ownership, purpose, and other relevant details.
    • Document AI use cases. Establish a process for reviewing and approving AI use cases. Consider what the technology is intended to accomplish, what information it can access, what actions it is permitted to take, and whether additional safeguards are necessary.
    • Review AI policies and training. Make sure policies address acceptable AI use, data handling, employee responsibilities, and the use of generative or agentic AI. Provide training so employees understand the institution's expectations and the risks associated with AI.
    • Conduct risk assessments. Evaluate AI use cases based on their potential impact, autonomy, data access, and operational dependencies. Use the results to determine appropriate controls and oversight.
    • Establish human oversight. Identify who is responsible for monitoring AI systems, reviewing their outputs or actions when necessary, and responding when issues arise. Consider whether employees have the authority and ability to intervene.
    • Evaluate third-party AI risks. Review how vendors use AI within their products and services. Consider data access, system permissions, underlying service providers, and how incidents are reported.
    • Keep the board and senior management informed. Provide appropriate reporting on material AI use cases, associated risks, significant incidents, and the effectiveness of risk management activities.
    • Review business continuity procedures. Consider how critical business processes would continue if an AI system became unavailable or unreliable. Identify alternate procedures and incorporate relevant AI-related disruptions into exercises and tests.
    • Monitor and test AI systems. Establish procedures for evaluating AI performance, identifying unexpected behavior, and reassessing risks when systems or use cases change. Consider whether existing controls remain effective as AI capabilities evolve.

The OCC's framework does not require financial institutions to establish an entirely new AI risk management program. Instead, institutions can use these considerations to evaluate how well their existing practices address AI-related risks and where additional controls may be appropriate.

How Tandem Supports AI Governance and Risk Management

Preparing for AI-related examiner questions may involve several areas of the institution, from cybersecurity and vendor management to policies and employee training. Tandem provides tools and resources that can help financial institutions bring those pieces together:

    • AI Cybersecurity Assessment Frameworks: Review AI-related cybersecurity controls using frameworks available in Tandem Cybersecurity Assessment.
    • AI Risk Assessment Template: Identify and evaluate risks associated with vendor-provided AI tools and services, including Microsoft Copilot, OpenAI ChatGPT, Anthropic Claude, and others in Tandem Risk Assessment.
    • AI Policy Template: Create clear expectations for acceptable AI use, data handling, third-party AI services, and oversight responsibilities with Tandem Policies.
    • AI Security Awareness Training Courses: Help employees understand AI-related risks, use AI responsibly, and protect organizational information with Tandem Phishing.
    • AI Vendor Review Templates: Evaluate how vendors use AI within their products, what data may be involved, and which controls are in place to manage the associated risks in Tandem Vendor Management.

Visit the Tandem AI Resources page for additional tools, educational content, and updates related to AI governance and risk management.

Frequently Asked Questions (FAQs)

What is the OCC Super Intelligence Supervisory Framework?
The OCC’s Super Intelligence: Supervisory Framework provides examiners with considerations for evaluating how banks manage artificial intelligence (AI) risks. It covers six pillars and 16 components, with particular attention to generative and agentic AI.

Does the OCC’s SI framework create new regulatory requirements for banks?
No. The framework supplements existing supervisory practices rather than establishing new regulatory requirements. Financial institutions can use it to better understand how examiners may evaluate their AI risk management practices.

Does the OCC require banks to have an AI risk management program?
The OCC does not require banks to establish a separate AI risk management program under this framework. Instead, banks can incorporate AI-related risk into existing governance, risk management, and compliance processes.

What is the difference between artificial intelligence (AI) and super intelligence (SI)?
The OCC uses the term Super Intelligence in place of Artificial Intelligence following Executive Order 14434. In this framework, the terminology change does not mean institutions are dealing with an entirely new type of technology.

How should banks assess risk associated with AI systems?
Banks can evaluate AI use cases based on factors such as autonomy, data access, dependencies, and potential customer or financial impact. The OCC provides an illustrative four-tier risk classification approach to help institutions consider controls appropriate to different levels of risk.

How does the OCC’s SI framework address agentic AI?
The framework places particular attention on AI systems that can independently plan, make decisions, and take actions. For these systems, it considers controls over what the system is permitted and authorized to do, how humans oversee and monitor its activity, and how its actions can be suspended or overridden.

How does the OCC’s SI framework address third-party risk management?
The framework encourages considerations of AI-related risks within existing third-party relationships. This includes understanding how vendors use AI, what data and systems their AI capabilities can access, and how changes or disruptions involving AI services may affect the bank.

How can banks prepare for OCC examinations involving AI?
Banks can review their AI inventories, document use cases, assess AI-related risks, evaluate existing policies and controls, and establish appropriate oversight. They can also review third-party relationships, provide employee training, and consider AI-related scenarios in business continuity planning.