Wealth management AI gateway SR 26-2 | Moring
The release of SR 26-2 reshapes how wealth management firms govern AI gateways by distinguishing between traditional quantitative models and emerging generative technologies. While the guidance narrows the definition of a model, it necessitates a parallel governance framework for AI agents. Firms must now balance compliance with SR 26-2 requirements while proactively managing the specific risks associated with autonomous AI systems.
Short answer
The release of SR 26-2 reshapes how wealth management firms govern AI gateways by distinguishing between traditional quantitative models and emerging generative technologies. While the guidance narrows the definition of a model, it necessitates a parallel governance framework for AI agents. Firms must now balance compliance with SR 26-2 requirements while proactively managing the specific risks associated with autonomous AI systems.
How it works
In the wealth management sector, the adoption of AI gateways for client service, investment research, and compliance operations has introduced significant governance challenges. The Federal Reserve's SR 26-2, issued on April 17, 2026, supersedes SR 11-7 and emphasizes a risk-based approach to model risk management [1, 5]. For wealth management leaders, the core challenge lies in the fact that while SR 26-2 narrows the definition of a model to complex quantitative methods, it expressly excludes generative and agentic AI from its scope [4, 8]. This creates a regulatory gap where firms must determine how to govern AI gateways that do not technically qualify as models under the new guidance [3].
To maintain control over AI gateways, firms must implement a parallel governance framework that addresses the specific risks of non-model AI systems. This involves assessing AI gateways based on reasoning complexity, degree of autonomy, and the potential impact of output errors [3]. By applying these dimensions, wealth management organizations can produce an actionable inventory that extends beyond traditional model classification [3]. This approach ensures that even when AI tools fall outside the formal SR 26-2 definition, they remain subject to rigorous oversight and internal controls that align with the spirit of sound model risk management principles [1, 4].
Regulators and industry experts emphasize that the exclusion of generative AI from SR 26-2 does not remove the obligation to govern these tools [3, 4]. FINRA notes that firms are increasingly using generative AI for summarization and information extraction, highlighting the need for supervisory processes to mitigate risks such as hallucinations and bias [2]. Furthermore, the lack of clarity regarding vendor models embedded with AI remains a critical concern for wealth management institutions, as many cannot fully inspect these third-party components [3, 7]. Proactive engagement with examiners is recommended, as explaining governance reasoning early can improve compliance processes and provide a strategic advantage [3].
For wealth management practitioners, the governance checklist for AI gateways should include several key actions. First, establish a defensible classification system that separates SR 26-2 models from generative and agentic AI agents [5]. Second, develop an inventory that captures sufficient information to understand risks, including vendor-provided components [8]. Third, implement kill-switch protocols and regulatory reporting mechanisms for AI failures, as these are identified as critical gaps in current architecture [7]. Finally, ensure that all AI outputs are faithful to approved sources, underlying model outputs, and accountable human decisions through a process of output fidelity [6].
Failure modes in AI gateway deployment are often linked to inadequate governance architecture rather than technical limitations. A significant risk involves the misinterpretation of policies or inaccurate client data generated by AI, which can impact decision-making in wealth management [2]. Furthermore, the reliance on skewed or limited training data can lead to concept drifts and biased outputs [2]. Without clear kill-switch protocols, firms face the risk of being unable to shut down malfunctioning AI models, as evidenced by industry surveys showing that many institutions are not confident in their incident response capabilities [7].
Wealth management firms should begin by assessing their current AI gateway landscape against the risk-based materiality framework. This involves mapping existing AI tools to the four-step functions of govern, map, measure, and manage [6]. By treating exam preparation as an opportunity to educate regulators on the firm's governance reasoning, wealth management leaders can foster constructive feedback and ensure their AI strategy remains compliant with the evolving expectations set forth by the agencies [3]. Establishing a transparent lifecycle for AI gateways, from initial evaluation to credible exit, is essential for maintaining trust and regulatory standing [5].
Frequently asked questions
How does SR 26-2 change the way wealth management firms classify AI gateways?
SR 26-2 narrows the definition of a model to complex quantitative methods that apply statistical, economic, or financial theories to process input data into quantitative estimates [8]. In wealth management, this means many AI gateways that operate on deterministic rule-based processes or simple calculations fall outside the scope of the guidance [4]. Consequently, firms must classify their AI gateways by distinguishing between those that qualify as models and those that function as generative or agentic AI, requiring a parallel governance framework for the latter [3].
Are wealth management firms exempt from governing AI gateways that fall outside the scope of SR 26-2?
No, the exclusion of generative and agentic AI from SR 26-2 does not remove the obligation to govern these systems [3]. Wealth management firms are still expected to manage risks associated with these AI gateways, such as accuracy, hallucinations, and bias [2]. Firms should adopt a risk-based materiality framework to assess these systems by their reasoning complexity, degree of autonomy, and the impact of potential output errors to ensure they meet internal and regulatory expectations [3].
What is the primary risk for wealth management firms using vendor-provided AI gateways?
The primary risk is the lack of transparency and control over vendor models embedded with AI, as most wealth management institutions cannot fully inspect these third-party components [3]. This creates unresolved scoping questions regarding who owns the governance obligation [3]. Because many firms lack confidence in their ability to shut down malfunctioning AI or report failures to regulators, they face significant operational and compliance risks when integrating these external AI gateways into their wealth management services [7].
How should wealth management firms prepare for exams regarding their AI gateway governance?
Wealth management firms should treat exam preparation as an opportunity to educate examiners on their governance reasoning [3]. By framing governance decisions around risk and demonstrating a clear, defensible classification system, firms can receive more constructive feedback [3]. It is essential to provide evidence of independent challenge, reliable evaluation, and traceable operation for every AI gateway, rather than relying on a policy statement alone to prove compliance with the principles of model risk management [5].