Book a demo
Topics · AI-CONTROL-PLANE

Asset management AI gateway SR 26-2 | Moring

Senior compliance and technology leaders in asset management navigating supervisory letter SR 26-2 must implement an ai gateway to govern generative models excluded from traditional model risk frameworks while maintaining provable compliance, centralized audit trails, and strict control over provider keys and agentic workflows.

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

Short answer

Senior compliance and technology leaders in asset management navigating supervisory letter SR 26-2 must implement an ai gateway to govern generative models excluded from traditional model risk frameworks while maintaining provable compliance, centralized audit trails, and strict control over provider keys and agentic workflows.

How it works

For senior compliance and technology leaders in asset management, the release of supervisory letter SR 26-2 by the federal reserve creates a distinct operational challenge regarding ai gateway deployments [1]. SR 26-2 narrows the traditional model definition to exclude generative and agentic artificial intelligence, which removes these advanced systems from standard model risk management frameworks while leaving institutional responsibility fully intact [2, 8]. This exclusion does not remove the obligation to govern these tools, requiring asset management firms to establish a parallel governance framework specifically for their ai gateway infrastructure [3, 6].

To manage this governance gap within asset management operations, an ai gateway functions as the mandatory network path and centralized control point through which all model traffic must pass before requests leave the firm perimeter [5]. By routing ai gateway requests centrally, asset management technology teams ensure that provider keys stop circulating among individual application developers and are instead replaced by scoped credentials [5]. This architecture allows compliance leaders to enforce deterministic role-based access control, manage token budgets, and capture comprehensive audit logs that satisfy institutional accountability requirements without relying on manual surveys of code repositories [5].

According to federal reserve supervisory guidance, SR 26-2 updates supervisory letter SR 11-7 and highlights sound principles for model risk management while recognizing that practices appropriately vary based on specific risk profiles and model usage [1]. KPMG notes that SR 26-2 largely codifies best practices from mature model risk management functions to accelerate strategic evolution, providing levers such as risk-based tiering and inventory de-scoping for excluded systems [2]. Moody's explains that governance must shift from the model alone to the full ai system, requiring a governance life cycle spanning design, validation, monitoring, escalation, and revalidation [3]. Furthermore, industry analysis indicates that nearly three in four financial professionals lack confidence in their kill-switch protocols or regulatory reporting for ai failures, underscoring the urgent need for robust gateway controls in asset management [7].

Implementing an effective ai gateway control environment in asset management requires a disciplined practitioner checklist based on regulatory expectations and industry frameworks. Asset management compliance teams must establish provable coverage where the ai gateway answers examiner inquiries with network paths rather than surveys [5]. Organizations must eliminate raw provider keys by issuing application teams gateway credentials scoped strictly to their operational permissions [5]. Centralized evidence collection must be maintained so that a single system holds the complete record of model calls and human policy modifications for audit reconstructions [5]. Finally, asset management firms must implement risk-based materiality assessments for ai systems based on reasoning complexity, degree of autonomy, and the impact of output errors [6].

Without an adequate ai gateway architecture, asset management institutions face severe failure modes during regulatory examinations and operational incidents. A primary failure mode is the inability to provide centralized evidence of what models did and what humans changed about the controls, which prevents effective reconstruction of audit trails [5]. Asset management firms also risk unmanaged vendor models embedded with third-party software where internal teams cannot fully inspect the underlying components or determine ownership of the governance obligation [6]. Additionally, failing to establish centralized gateway controls leaves organizations vulnerable when developer turnover occurs, as locally written client wrappers and policies disappear along with the engineers who created them [5].

Asset management compliance and technology teams should start their transformation by mapping all existing generative and agentic workloads through a centralized ai gateway deployment pattern [5]. Teams should engage proactively with examiners by explaining their governance reasoning around risk-based materiality, treating examination preparation as an opportunity to educate stakeholders [6]. Leadership must establish a parallel governance framework for systems excluded from traditional model risk management, focusing on correctness, consistency, robustness, and predictability across the foundation-model backbone and its surrounding harness [3]. Establishing these foundational controls ensures that asset management firms can safely scale artificial intelligence adoption while maintaining rigorous oversight.

Frequently asked questions

How does supervisory letter SR 26-2 impact generative artificial intelligence usage within asset management?

Supervisory letter SR 26-2 specifically narrows the model definition to exclude generative and agentic artificial intelligence from traditional model risk frameworks. For asset management firms, this regulatory change means that while these advanced systems are carved out of standard model definitions, the institution retains full responsibility for governing them through an ai gateway and parallel control frameworks.

Why must asset management firms route model traffic through a centralized ai gateway?

Routing model traffic through a centralized ai gateway allows asset management organizations to enforce provable coverage, stop raw provider keys from circulating among developers, and centralize audit logs. An ai gateway ensures that compliance leaders can answer examiner inquiries with verifiable network paths and maintain strict governance over all ai interactions within asset management operations.

What specific audit evidence must asset management compliance leaders capture via their ai gateway?

Asset management compliance leaders must capture two distinct layers of audit evidence through their ai gateway: records detailing what the models actually did during execution, and records tracking what humans changed regarding the underlying controls. Centralizing this evidence via an ai gateway enables asset management firms to satisfy comprehensive examination requirements.

How can asset management technology teams address vendor models embedded with artificial intelligence under SR 26-2?

Asset management technology teams facing unresolved scoping questions around vendor models embedded with artificial intelligence must establish robust risk-based materiality assessments through their ai gateway. Because institutions cannot fully inspect deeply embedded third-party components, implementing gateway-level controls and proactive monitoring helps asset management firms manage unobservable risks.

Sources

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