Credit unions AI agent registry SR 11 7 | Moring
Credit unions must maintain an AI agent registry to satisfy examiner expectations regarding model risk management. By documenting every agent that interacts with member data or lending decisions, credit unions demonstrate clear ownership, risk assessment, and operational control. This registry serves as the primary evidence of governance maturity, ensuring compliance with SR 11-7 principles while managing the risks inherent in agentic workflows.
Short answer
Credit unions must maintain an AI agent registry to satisfy examiner expectations regarding model risk management. By documenting every agent that interacts with member data or lending decisions, credit unions demonstrate clear ownership, risk assessment, and operational control. This registry serves as the primary evidence of governance maturity, ensuring compliance with SR 11-7 principles while managing the risks inherent in agentic workflows.
How it works
For credit unions, the transition from traditional model risk management to governing AI agents requires a shift in how institutions document and monitor automated systems. While SR 11-7 established the foundation for model risk management, the rise of agentic AI creates a need for an AI agent registry that tracks systems touching member data or decisioning workflows. Credit unions that maintain this registry open their examinations with credibility, whereas those attempting to assemble lists during an exam often face findings regarding oversight and transparency.
The control mechanism relies on a structured inventory that bridges the gap between technical deployment and regulatory accountability. A robust registry identifies the specific agents in use, their purpose, the data they access, and the human oversight mechanisms governing their actions. By mapping these agents, credit unions can define clear decision authority, ensuring that for lending processes, a human officer remains the final decision-maker while the agent merely assembles the file. This approach aligns with the cooperative principles and member-focused mission of credit unions by ensuring that automated tools remain subservient to institutional policy.
Regulatory guidance, including SR 26-2 and the foundational SR 11-7, emphasizes that institutions must maintain sufficient information to understand model risks. Examiners expect credit unions to provide a comprehensive inventory of AI systems and agents that influence member decisions. While SR 26-2 notes that generative and agentic AI are outside its specific scope, it mandates that an institution’s risk management and governance should guide controls for these technologies. Consequently, credit unions must treat the registry as a critical component of their broader model risk management framework, ensuring that even non-model agents are subject to documented oversight.
To build an effective registry, credit unions should follow a structured checklist: First, inventory every AI use case, naming the owner, purpose, and data limitations. Second, assign a risk rating based on the institution's established policy. Third, document validation reports or testing results before the agent's first use. Fourth, explicitly define human review requirements for agent outputs. Fifth, establish ongoing performance monitoring against expected outcomes. Finally, maintain records of third-party oversight, including vendor contracts, audit rights, and model-change notifications for any agent provided by an external partner.
Failure modes often arise when credit unions neglect the registry, leading to silent drift in agent behavior or unauthorized tool abuse. Without a clear behavioral baseline defined in the registry, agents may pursue technically valid but unintended goals, such as prompt injection or data leakage. Furthermore, failing to document ownership and decision authority can lead to regulatory findings if an examiner cannot trace an adverse action to a specific human decision record. Managing these risks requires proactive financial management, including tracking token consumption and compute costs, to prevent budget overruns and ensure only authorized personnel manage deployments.
Credit unions should begin by establishing a unified view of agent usage across departments. This involves auto-instrumenting agents to track performance metrics consistently and applying cost center tags to allocate expenses accurately. By integrating these governance signals into a central registry, credit unions can set real-time alerts to notify teams of budget thresholds or performance deviations. This structured approach allows credit unions to scale AI adoption safely while maintaining the rigorous governance standards expected by examiners under the principles of SR 11-7 and SR 26-2.
Frequently asked questions
Why is an AI agent registry necessary for credit unions under SR 11-7?
Under the model risk management principles of SR 11-7, credit unions are expected to maintain an inventory of all systems that impact member decisions. An AI agent registry provides the necessary evidence that the credit union understands its risk exposure, has assigned clear ownership, and maintains human oversight over automated agentic workflows.
How does an AI agent registry assist during a regulatory examination?
When examiners ask for a list of AI systems touching member data, a maintained AI agent registry allows a credit union to provide an immediate, accurate response. This credibility prevents the findings that typically occur when credit unions struggle to assemble documentation during the exam process, ensuring compliance with SR 26-2 expectations.
What should be included in the AI agent registry for third-party tools?
For credit unions, the registry must include vendor contracts, data-use terms, and audit rights for any third-party agent. Because SR 11-7 requires the same validation rigor for vendor models as in-house ones, the registry should also store validation reports, model certifications, and any relevant SOC reports to demonstrate thorough third-party oversight.
How do credit unions define human review in the AI agent registry?
The registry must document exactly which agent outputs require human approval before execution. For credit unions, this means ensuring that for lending decisions, a human officer retains final authority, while the AI agent registry records the specific reviewer and the decision-making process to ensure transparency and accountability for all member-facing actions.