Book a demo
Topics · MCP

Wealth management MCP governance SR 26-2 | Moring

SR 26-2 introduces a risk-based approach to model risk management that requires wealth management firms to calibrate governance to their specific complexity. While the guidance narrows the definition of a model to exclude generative and agentic AI, senior leaders must still apply robust MCP governance to these tools to mitigate risks like hallucinations and bias.

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

Short answer

SR 26-2 introduces a risk-based approach to model risk management that requires wealth management firms to calibrate governance to their specific complexity. While the guidance narrows the definition of a model to exclude generative and agentic AI, senior leaders must still apply robust MCP governance to these tools to mitigate risks like hallucinations and bias.

How it works

For wealth management leaders, the transition from SR 11-7 to SR 26-2 necessitates a fundamental shift in how MCP governance is applied to complex quantitative systems. As wealth management firms increasingly leverage sophisticated models for client advisory and portfolio construction, the regulatory focus has moved toward a risk-based approach that scales with a firm's specific complexity and model usage [1]. This means that the depth of oversight required for a high-frequency trading model differs from that of a standard client reporting tool, requiring a more nuanced application of model risk management principles [1].

Effective MCP governance under this new guidance functions by retaining the core conceptual framework of its predecessor—including model definition, inventory, independent validation, and effective challenge—but removing the rigid, prescriptive techniques and role lists that previously defined the process [3, 7]. Instead, the framework mandates that the intensity of validation and monitoring be directly proportional to the model's risk profile [1]. In the context of wealth management, this requires maintaining a rigorous model inventory and ensuring that the three-lines-of-defense model—comprising model owners, MRM/compliance, and internal audit—is calibrated to the firm's specific size and operational complexity [3, 6].

Regulatory guidance and research indicate that SR 26-2 explicitly narrows the definition of a model to exclude generative AI and agentic AI, which are not considered complex quantitative methods applying statistical or financial theories in the same way [7]. However, sources emphasize that this exclusion does not remove the obligation to govern these tools; rather, banking organizations' existing risk management and governance practices should guide the determination of appropriate controls for any tools or processes used [2]. While SR 26-2 provides the framework for traditional models, practitioners must develop parallel frameworks for AI-driven tools to address risks such as hallucinations, where models generate inaccurate information, and bias, where outputs are skewed by limited or outdated training data [4, 5].

A practitioner checklist for wealth management MCP governance should include: maintaining a comprehensive model inventory with sufficient information to understand specific risks [7]; establishing a process for independent validation and effective challenge by qualified personnel not involved in development [3, 6]; implementing ongoing monitoring and outcomes analysis to detect concept drift [3, 7]; and ensuring robust books-and-records retention for all validation memos and model artifacts [6]. Additionally, leaders should assess AI systems based on reasoning complexity, degree of autonomy, and the potential impact of output errors to create an actionable inventory [5].

Failure modes in wealth management MCP governance often stem from a lack of clarity regarding third-party tools. Because SR 26-2 does not resolve who owns the governance obligation for AI components embedded within vendor models, firms face significant exposure when they cannot fully inspect the underlying logic of third-party software [5]. Furthermore, a lack of preparedness regarding AI failure protocols is a critical vulnerability; research indicates that many institutions lack the ability to implement kill-switch protocols or provide regulatory reporting when an AI model fails [8]. Such failures can lead to significant legal consequences, including settlements related to discriminatory outputs caused by models using illegal proxies for decision-making [8].

To begin, wealth management teams should engage in proactive examiner engagement to explain their governance reasoning early, which can lead to more constructive feedback [5]. Teams should start by mapping their existing model inventory against the new risk-based criteria and identifying where generative AI or agentic AI tools currently operate outside the formal model definition [5, 7]. By framing governance decisions around risk and treating the transition as an opportunity to refine the three-lines-of-defense structure, firms can build a more resilient MCP governance framework that prepares them for future, more specific AI-related guidance [5].

Frequently asked questions

How does SR 26-2 change the way wealth management firms define a model?

SR 26-2 narrows the definition of a model to focus on complex quantitative methods that apply statistical, economic, or financial theories to process data [7]. Crucially, this definition excludes generative AI and agentic AI, meaning wealth management MCP governance must now account for these tools through separate, risk-based governance practices rather than traditional model risk management alone [2, 7].

What are the primary risks of using generative AI in wealth management under current guidance?

While generative AI is not formally classified as a model under SR 26-2, wealth management firms face risks including hallucinations, where the tool presents inaccurate information as factual, and bias, where outputs are skewed by limited or outdated training data [4]. Effective MCP governance is required to identify and mitigate these specific risks to ensure decision-making remains accurate [4].

How should wealth management leaders approach third-party AI vendor models?

Managing third-party risk is a significant challenge because SR 26-2 does not clearly define who owns the governance obligation for AI components embedded within vendor software [5]. Wealth management firms must develop robust MCP governance to address these unresolved scoping questions, especially since many institutions cannot fully inspect the internal logic of these embedded AI systems [5].

What is the importance of the three-lines-of-defense in wealth management MCP governance?

The three-lines-of-defense model remains a core component of the SR 26-2 framework, involving model owners in the first line, MRM and compliance in the second, and internal audit in the third [3, 6]. For wealth management, this structure ensures that MCP governance is applied consistently, providing independent validation and effective challenge to maintain the integrity of complex quantitative processes [3, 6].

Sources

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