Book a demo
Topics · REGULATORY

Wealth management AI model risk management SR 11 7 | Moring

Wealth management firms must evolve model risk management to address the shift from deterministic systems to generative and agentic AI. While SR 26-2 replaces SR 11-7, the core requirement for independent validation and governance remains critical. Firms must now manage unique AI risks like model drift, hallucinations, and vendor opacity to maintain institutional safety and soundness.

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

Short answer

Wealth management firms must evolve model risk management to address the shift from deterministic systems to generative and agentic AI. While SR 26-2 replaces SR 11-7, the core requirement for independent validation and governance remains critical. Firms must now manage unique AI risks like model drift, hallucinations, and vendor opacity to maintain institutional safety and soundness.

How it works

In wealth management, the integration of AI models introduces complexities that traditional risk frameworks struggle to contain. While SR 11-7 defined models as systems applying statistical or financial theories to process data, modern wealth management platforms now incorporate generative and agentic AI. These systems often operate in a governance gap, as their non-deterministic nature and emergent behaviors differ significantly from the stable, inspectable logic of traditional quantitative models. For wealth management leaders, the challenge is ensuring that these advanced tools remain subject to the same rigor of oversight as legacy valuation models.

Effective control in this environment requires embedding SR 11-7 principles into firm policies while adapting them for AI-specific features. This involves maintaining a comprehensive model inventory, assigning clear roles, and ensuring independent model review. Because wealth management firms often rely on third-party AI services, they must apply validation rigor to vendor models even when underlying code or data access is limited. The separation of responsibility remains paramount; the individuals responsible for building a wealth management model should not be the same parties conducting its validation.

Regulators have shifted the landscape with the April 2026 release of SR 26-2, which supersedes SR 11-7. While SR 26-2 notes that the guidance does not set forth prescriptive, enforceable standards—meaning non-compliance alone may not trigger supervisory criticism—safety and soundness issues resulting from poor model risk management practices remain firmly in scope. Sources indicate that wealth management firms must now account for model drift, explainability, and the unique failure modes of generative AI, such as hallucinations or prompt-dependent outputs that lack historical analogs in traditional financial modeling.

Practitioners in wealth management should follow a structured checklist to maintain compliance: perform explainability testing on all AI outputs, conduct scenario-based stress testing for generative models, and implement tiering frameworks that classify models by materiality and complexity. It is essential to request validation reports, model certifications, and SOC reports from vendors upfront to avoid delays during examinations. Furthermore, wealth management teams must distinguish between simple rule-based processes, which fall outside this guidance, and complex quantitative systems that require formal validation.

Failure modes in AI model risk management often stem from the transition to high-dimensional feature spaces and opaque learned functions. Unlike a logistic regression, where coefficients are readable, modern AI models in wealth management can act as moving targets due to frequent retraining and feature drift. Misuse of a fundamentally valid model also poses significant risk; decision makers must fully understand the limitations and objectives of their AI tools to avoid outcomes that undermine stakeholder trust or lead to misleading financial outputs.

To begin, wealth management firms should formalize their governance by identifying all sources of model risk and assessing their aggregate impact. This requires establishing an effective challenge process where technically competent parties critically analyze models against their stated objectives. By focusing on the separation of duties and maintaining evidence of compliance, firms can bridge the gap between traditional SR 11-7 expectations and the requirements of the modern AI-driven wealth management landscape.

Frequently asked questions

How does SR 26-2 change the approach to AI model risk management for wealth management firms?

SR 26-2 supersedes SR 11-7 and clarifies that the guidance does not establish prescriptive, enforceable standards, which may reduce the threat of immediate supervisory criticism. However, for wealth management firms, safety and soundness remain the priority. AI model risk management must now account for the unique challenges of generative AI, such as non-deterministic outputs and the governance gap created by agentic systems that orchestrate traditional models.

What constitutes a model under the current guidance for wealth management firms?

A model is defined as a complex quantitative method, system, or approach that applies statistical, economic, or financial theories to process input data into quantitative estimates. In wealth management, this excludes simple arithmetic calculations or deterministic rule-based software. AI model risk management must focus on models that rely on assumptions and complex logic, as these represent the core risk to institutional stability.

Why is independent validation critical for wealth management AI models?

Independent validation ensures that models function as intended and align with wealth management objectives. Because developers may have inherent biases, separating the building and validation roles is essential. In AI model risk management, this process involves critical analysis by informed parties who can identify limitations, assumptions, and potential failure modes that the original developers might overlook during the initial design phase.

How should wealth management firms manage risks associated with third-party AI models?

Wealth management firms must apply the same validation rigor to third-party models as they do to in-house systems. Since full access to vendor code is often restricted, firms should proactively request validation reports, model certifications, and SOC reports. Effective AI model risk management requires these firms to evaluate the vendor's underlying assumptions and performance data to ensure the model does not introduce systemic instability.

Sources

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