Book a demo
Topics · REGULATORY

Asset management AI model risk management SR 11 7 | Moring

Asset management leaders must navigate the transition from SR 11-7 to SR 26-2 to ensure robust ai model risk management. While SR 11-7 established foundational governance and validation standards, the new guidance narrows the definition of a model to exclude generative and agentic AI, requiring specialized oversight for these evolving technologies.

On this page
  1. Short answer
  2. How it works
  3. FAQs
  4. Related reading
  5. Sources

Short answer

Asset management leaders must navigate the transition from SR 11-7 to SR 26-2 to ensure robust ai model risk management. While SR 11-7 established foundational governance and validation standards, the new guidance narrows the definition of a model to exclude generative and agentic AI, requiring specialized oversight for these evolving technologies.

How it works

For asset management firms, adhering to SR 11-7 is critical because model risk can undermine the validity of quantitative outputs and investment decisions. As firms integrate complex mathematical calculations and theoretical frameworks into their operations, the potential for fundamental errors during design or implementation increases. Effective ai model risk management ensures that these models do not produce inaccurate results that could lead to significant financial or operational consequences for the institution [3].

Control within this framework functions through rigorous governance, policies, and independent validation. Under SR 11-7, banks are expected to maintain a comprehensive model inventory and assign clear roles to ensure accountability [2]. A key component is the separation of duties, where the individuals who build a model are distinct from those who validate it to ensure objective judgment [2]. This separation of responsibility is more vital than the specific organizational chart, especially for smaller teams where a designated reviewer with real authority may suffice [2].

Regulators and cited sources highlight a shifting landscape. While SR 11-7 provides guidance for models supporting significant business lines, it does not set enforceable standards or prescriptive requirements [1]. However, the introduction of SR 26-2 changes the scope by narrowing the definition of a model to exclude generative and agentic AI [4]. This means that while existing documentation, validation reports, and monitoring reports remain necessary, the rationale for their application is changing [4]. Furthermore, traditional validation methods struggle with machine learning and generative AI because these technologies often lack the deterministic logic and inspectable functions found in traditional models [5].

To maintain compliance, practitioners should utilize the following checklist: maintain a firm-wide model inventory with sufficient information to understand risks [4]; ensure independent validation is performed by technically competent parties capable of effective challenge [3]; secure validation reports, model certifications, and SOC reports for all third-party or vendor models [2]; and establish clear documentation standards for both development and implementation [4].

Failure modes in ai model risk management can be catastrophic. Fundamental errors may arise from incorrect design choices or inappropriate inputs, leading to downstream consequences [3]. Models may also be misused if decision makers do not understand their inherent limitations [3]. Additionally, machine learning models can suffer from feature drift and opaque learned functions, making them moving targets that resist traditional line-by-line review [5]. Generative AI introduces further risks through non-deterministic outputs and emergent behaviors that traditional pre-enumerated test cases cannot easily capture [5].

Teams should begin by assessing their current model inventory and identifying which models fall under the traditional SR 11-7 definition versus those that are now categorized as generative or agentic AI under SR 26-2 [4]. Organizations must also evaluate their third-party model risks by requesting vendor validation reports upfront to avoid delays during regulatory exams [2]. Establishing a culture of effective challenge, where informed parties critically analyze models against their objectives, is the most vital starting point for managing aggregate model risk [3].

Frequently asked questions

How does the transition from SR 11-7 to SR 26-2 affect asset management oversight?

The transition requires asset management firms to distinguish between traditional quantitative models and newer technologies. While SR 11-7 provided a framework for managing model risk through governance and validation, SR 26-2 narrows the definition of a model to exclude generative and agentic AI, necessitating a different approach to ai model risk management for these specific tools.

What are the specific challenges of validating third-party models in asset management?

In asset management, validating third-party models is difficult because firms often lack full access to the underlying code or data. To maintain effective ai model risk management, practitioners should request vendor validation reports, model certifications, and SOC reports early to ensure that the vendor's model meets the same rigor as in-house developments.

Why is 'effective challenge' important for ai model risk management?

Effective challenge is critical in asset management because it involves technically competent parties performing a critical analysis of a model against its objectives. This process helps identify model limitations and assumptions, which is essential for ai model risk management as machine learning and generative AI can exhibit non-deterministic behaviors and opaque logic.

What risks are associated with the misuse of models in asset management?

Even a fundamentally valid model can pose high risk in asset management if it is misused. Decision makers must understand the limitations of every model to avoid using them in ways inconsistent with their original intent, a core principle of ai model risk management that prevents errors in quantitative estimates and downstream financial consequences.

Sources

  1. - federalreserve.gov
  2. - federalreserve.gov
  3. - artificialintelligenceact.eu