The Office of the Comptroller of the Currency (OCC) has introduced its first supervisory framework for artificial intelligence in the U.S. banking industry, establishing new guidance for evaluating the risks associated with autonomous AI agents handling financial transactions. Announced on November 4, 2026, the framework arrives as major financial institutions, including Morgan Stanley, JPMorgan Chase, Bank of America and Citi, restrict third-party AI agents from accessing customer accounts amid concerns that autonomous software could execute transactions beyond users’ instructions. The OCC’s 22-page guidance emphasizes financial consequences, human oversight and operational safeguards rather than the sophistication of individual AI models. Although the framework does not establish legally binding regulations, it represents a significant step toward addressing the growing cybersecurity and financial stability risks associated with agentic AI in banking.
OCC Introduces First Federal Supervisory Framework for Banking AI
On November 4, the Office of the Comptroller of the Currency, one of the primary regulators overseeing national banks in the United States, released its first framework specifically addressing artificial intelligence and autonomous AI systems.
The guidance is designed to help federal bank examiners assess how financial institutions deploy AI technology across banking operations, including payments, lending, fraud detection, customer service and cybersecurity.
Unlike traditional banking software that operates according to predetermined instructions, agentic AI systems can independently perform multistep tasks, interact with external applications and make operational decisions with varying degrees of human supervision.
These capabilities create new opportunities for financial institutions to improve efficiency but also introduce risks when AI systems gain access to sensitive customer information or the authority to move money.
The OCC’s framework directs examiners to evaluate AI applications based on their potential consequences rather than simply assessing how technologically advanced the underlying models are.
For example, a relatively basic AI agent authorized to initiate bank transfers could present greater financial risk than a sophisticated AI model used exclusively to summarize internal documents.
The agency emphasized that banks should manage these technologies within their existing operational, compliance and cybersecurity risk frameworks rather than treating artificial intelligence as an entirely separate category.
Morgan Stanley and JPMorgan Restrict Unauthorized AI Agents
The regulatory announcement follows reports that several of America’s largest financial institutions have discovered third-party AI agents attempting to operate inside customer banking accounts.
An October 22 report revealed that Morgan Stanley, JPMorgan Chase, Bank of America and Citi were taking steps to restrict external AI systems that accessed banking platforms using customer credentials.
Some of these agents reportedly attempted actions that appeared to extend beyond the instructions provided by account holders.
The situation has raised concerns about whether banks can reliably distinguish between software legitimately authorized by customers and autonomous agents operating outside their intended permissions.
Morgan Stanley has blocked third-party AI agents from accessing its platforms, although the bank reportedly had not identified confirmed cases of unauthorized or rogue agents operating within its systems.
JPMorgan Chase, meanwhile, has emphasized the importance of controlling which external AI systems can interact with customer accounts and internal banking infrastructure.
Other institutions, including Bank of America and Citi, have implemented restrictions intended to limit unauthorized access.
The concern is particularly significant because some customers have reportedly provided their banking login credentials directly to AI assistants, potentially allowing those systems to interact with accounts as though they were the account holders.
This creates a difficult security challenge because transactions initiated through valid customer credentials may not immediately appear unauthorized to traditional fraud detection systems.
OCC Classifies Autonomous Payment Agents as High-Risk Technology
One of the most important components of the new framework is its approach to evaluating different levels of AI-related financial risk.
The OCC outlines four illustrative risk tiers, allowing examiners to consider how much independence an AI system possesses and the potential damage it could cause if something goes wrong.
Rather than applying the same oversight requirements to every AI application, the framework encourages a risk-based approach.
Examples of the distinctions include:
Highest risk: Fully autonomous payment agents and real-time credit decision systems capable of producing significant financial consequences.
Elevated risk: AI systems with substantial operational authority or access to sensitive financial functions.
Lower risk: AI applications operating under meaningful human supervision with limited decision-making authority.
Lowest risk: Internal productivity tools performing tasks such as document drafting or administrative assistance.
The agency recommends that financial institutions consider whether an AI-generated action can be reversed, how much authority the system possesses and whether human operators can intervene before serious financial consequences occur.
An AI assistant that helps employees prepare reports may require fewer safeguards than an autonomous system authorized to transfer millions of dollars between accounts.
The distinction reflects growing regulatory concern that AI systems capable of taking independent actions can introduce risks beyond those associated with conventional software.
Banks Urged to Implement AI Kill Switches and Transaction Limits
The OCC’s framework addresses the possibility that autonomous AI agents could continue operating even after making harmful decisions or exceeding their assigned permissions.
To reduce these risks, the guidance identifies several safeguards that bank examiners can consider when evaluating AI-powered financial systems.
These include transaction limits, identity verification, delegated permissions, continuous monitoring and emergency shutdown controls.
A particularly important recommendation involves establishing mechanisms that allow financial institutions to immediately suspend or revoke an AI agent’s access.
These controls, sometimes described as AI kill switches, could prevent unauthorized transactions from continuing after suspicious activity is detected.
The framework also highlights the importance of restricting AI systems to clearly defined operational responsibilities.
For example, a customer might authorize an AI assistant to review account balances, analyze spending patterns or recommend financial decisions without granting permission to transfer funds.
Banks could use restricted permissions to separate informational access from transaction execution, reducing the potential consequences of compromised or malfunctioning AI systems.
Additional safeguards may include:
Human approval: Requiring employee or customer confirmation before high-value or irreversible transactions.
Rate limits: Restricting how frequently an autonomous system can execute financial actions.
Access revocation: Immediately disabling compromised or unauthorized AI agents.
Independent testing: Evaluating AI systems for unexpected behavior before deploying them in sensitive banking operations.
These measures are intended to allow financial institutions to benefit from AI automation while maintaining control over activities that could affect customer assets or banking operations.
JPMorgan CEO Jamie Dimon Warns Against Unrestricted AI Access
The banking industry’s concerns have also been reflected in public statements from senior financial executives.
JPMorgan Chase CEO Jamie Dimon recently emphasized that the bank would not permit unrestricted access by outside AI agents because of the potential risks to customers and the institution.
The comments reflect a broader debate over how banks should accommodate customers who increasingly want AI assistants to manage their financial activities.
While autonomous agents could eventually help consumers pay bills, move funds, monitor investments and manage financial accounts, granting these systems unrestricted access creates significant security and accountability concerns.
Charles Schwab CEO Richard Wurster has also acknowledged that his company actively blocks unauthorized AI agents attempting to access its financial platforms.
Wurster said the brokerage monitors attempted access by external agents and maintains restrictions to protect authentication procedures and customer interactions.
Some financial institutions are considering read-only access as a potential compromise.
Under this approach, AI assistants could retrieve account information and analyze financial data without receiving permission to execute transactions.
This would resemble existing integrations used by budgeting applications and financial management platforms.
However, broader adoption of autonomous financial agents will likely require more sophisticated authorization systems capable of verifying user consent and limiting the actions software can perform.
Federal Regulators Increase Scrutiny of AI Across Financial Institutions
The OCC’s announcement follows increased attention from other U.S. financial regulators, including the Federal Reserve and Federal Deposit Insurance Corporation (FDIC).
Regulators have been asking banks how they use advanced AI systems, particularly in areas involving cybersecurity, financial transactions and operational decision-making.
Their inquiries have focused on whether institutions can effectively monitor autonomous agents, identify unexpected behavior and intervene when systems operate outside their intended responsibilities.
The growing scrutiny reflects concerns that AI-driven mistakes or cyberattacks could create consequences extending beyond individual financial institutions.
For example, an autonomous system responsible for executing large financial transactions could potentially create liquidity problems or operational disruptions if it began making unauthorized decisions.
Regulators are also examining the risks associated with third-party AI technology providers, particularly when multiple financial institutions rely on the same external systems.
A widespread failure involving a commonly used AI service could affect several banks simultaneously, potentially creating broader financial stability concerns.
These developments demonstrate why regulators are increasingly examining the relationship between artificial intelligence, cybersecurity and the resilience of financial infrastructure.
OCC Guidance Does Not Create New Banking Regulations
Despite its significance, the OCC emphasized that the newly released framework does not establish additional legally enforceable requirements.
Instead, it provides supervisory guidance that examiners can use when evaluating how banks manage AI-related risks under existing banking laws and regulatory expectations.
Financial institutions are not required to create entirely separate AI risk management departments or establish a distinct regulatory classification for artificial intelligence.
The framework also does not prohibit banks from deploying autonomous AI agents or developing AI-powered financial services.
Rather, it encourages institutions to implement controls appropriate to the potential risks associated with specific applications.
The OCC has indicated that it intends to update the framework as AI technology continues evolving and new operational challenges emerge.
Although the guidance is nonbinding, banks generally take supervisory expectations seriously because regulatory examinations can influence how institutions manage risk and demonstrate compliance.
The agency’s approach suggests that federal regulators are attempting to balance financial innovation with the need to protect customers and maintain banking system stability.
Agentic AI Could Transform Banking and Cryptocurrency Payments
The new framework arrives as agentic AI becomes increasingly relevant to both traditional finance and the cryptocurrency industry.
AI-powered financial agents are being developed to perform tasks such as monitoring investments, executing trades, managing payments and interacting with blockchain applications.
In cryptocurrency markets, autonomous agents can potentially interact with smart contracts, decentralized exchanges and digital wallets without requiring human approval for every individual action.
These capabilities could support a new generation of automated financial services, particularly as stablecoins and blockchain-based payment infrastructure become more widely adopted.
However, autonomous cryptocurrency transactions introduce additional concerns because blockchain transfers are generally irreversible once finalized.
An AI agent that mistakenly transfers cryptocurrency to an incorrect address or interacts with a malicious smart contract may create losses that cannot easily be recovered.
The OCC’s emphasis on transaction limits, delegated permissions and human oversight could therefore have implications beyond conventional banking.
Although the framework specifically addresses supervised banking institutions, its risk-management principles may also influence how cryptocurrency companies design autonomous payment systems and AI-powered wallet infrastructure.
As financial institutions increasingly explore stablecoin payments and tokenized assets, establishing clear boundaries for AI-controlled transactions could become an important component of digital asset security.
New AI Framework Signals Shift Toward Regulated Autonomous Finance
The OCC’s first AI supervisory framework represents an important milestone in the evolution of artificial intelligence within the U.S. financial system.
Rather than attempting to regulate AI solely according to model capabilities, the agency is emphasizing the real-world consequences of allowing autonomous systems to make financial decisions.
The approach recognizes that even relatively simple AI applications can create significant risks when connected to bank accounts, payment networks or sensitive customer information.
For financial institutions, the challenge will be developing systems that allow AI assistants to improve customer experiences without granting them excessive control over financial assets.
For technology companies building autonomous financial agents, the framework highlights the importance of transparent authorization, limited permissions and reliable emergency controls.
The growing restrictions imposed by Morgan Stanley, JPMorgan and other major banks demonstrate that the financial industry is already confronting these challenges as customers begin connecting AI agents to their accounts.
As artificial intelligence becomes more deeply integrated into banking, cryptocurrency trading and digital payments, regulators will face increasing pressure to establish safeguards that protect consumers without preventing innovation.
The OCC’s framework offers an early indication of how that balance may develop, placing human accountability, financial security and operational control at the center of the emerging autonomous finance ecosystem.
