Book a demo
Topics · REGULATORY

Asset Management Model Risk Roadmap | Moring

A model risk roadmap for asset management converts governance intent into dated milestones: inventory existing models and agents, define ownership and tiering, stand up monitoring and evidence pipelines, gate production deployments, then institutionalize quarterly review and vendor oversight. Research recommends discovery before policy, and sequencing controls where they reduce risk fastest.

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

Short answer

A model risk roadmap for asset management converts governance intent into dated milestones: inventory existing models and agents, define ownership and tiering, stand up monitoring and evidence pipelines, gate production deployments, then institutionalize quarterly review and vendor oversight. Research recommends discovery before policy, and sequencing controls where they reduce risk fastest.

How it works

A roadmap exists because model risk cannot be governed in a single program meeting. Industry research on AI governance roadmaps describes converting stated principles into dated milestones, named owners, and retrievable evidence across the first year of adoption, with a clear sequencing principle: discovery comes before policy, because governance cannot control models, vendors, or agents it has not found.

For an asset manager, discovery means a living inventory of everything that behaves like a model or agent: factor and risk models, valuation tools, research assistants, client reporting generators, and the automation connecting them. Governance research stresses the living inventory as one of the controls that actually reduce model risk, and lifecycle research observes that 97 percent of breached organizations lacked AI access controls, so discovery should record access scopes alongside ownership.

The second phase assigns structure. AI-MRM research describes tiering, drift detection, and attestation as the controls extending SR 11-7 principles to generative and agentic AI, so the roadmap should set model tiers by decision impact, name owners, and define agent authority as asset management governance research recommends for extending existing compliance principles to AI.

The third phase makes controls run. Platform research describes centralized layers that track models, score risk, and enforce policies, and governance research contrasts policy documents with self-generating evidence. Milestones here include live monitoring dashboards, immutable logging of model calls and agent actions, permission enforcement at runtime, and rollback paths tested before they are needed.

The regulatory dates give the calendar teeth. SR 26-2, adopted April 17, 2026 by the Federal Reserve, the OCC, and the FDIC, replaced SR 11-7 in the first model risk overhaul in fifteen years, and commentary notes it regulates models rather than agents, so agent milestones are firm-designed. EU AI Act obligations supply additional dated duties for firms with European operations.

A realistic year-one sequence for an asset manager: quarter one, inventory and tiering; quarter two, ownership, authority definitions, and monitoring for the highest tiers; quarter three, runtime enforcement and evidence pipelines across the estate; quarter four, quarterly review institutionalized with vendor oversight and exception management. Sequenced this way, the roadmap produces visible risk reduction early while building toward the complete evidence posture supervisors expect.

Two sequencing cautions come from the research. First, evidence should switch on with the first milestone, not after the program matures, because attestation and review depend on records that only exist if logging was built early. Second, vendor models and external agents belong in discovery from day one, since lifecycle research shows externally hosted models can change behind stable interfaces, and a roadmap that discovers them late inherits an ungoverned exposure retroactively.

Frequently asked questions

What should come first in a model risk roadmap?

Discovery. Governance research is explicit that an AI governance roadmap cannot control models, vendors, or agents it has not found, so a living inventory of every model, agent, and access scope comes first. For asset managers this spans factor models, valuation tools, research assistants, and reporting generators alike.

How should milestones be sequenced?

After discovery: tiering and ownership, then monitoring and evidence pipelines for the highest tiers, then runtime enforcement of agent permissions and limits, then institutionalized quarterly review and vendor oversight. Each asset management milestone should carry a date, an owner, and retrievable evidence as its definition of done.

Why does the roadmap need dated milestones at all?

Because principles without dates do not constrain anything. Roadmap research converts principles into dated milestones, named owners, and retrievable evidence, and governance research notes that model risk is reduced by mechanisms rather than documents. Dates also align the program with SR 26-2 and EU AI Act obligations.

How do agents fit into the roadmap?

As first-class inventory items with authority definitions, access scopes, action logging, and tiering of their own. Asset management governance research recommends extending existing compliance principles by defining agent authority, and commentary on SR 26-2 notes it regulates models rather than agents, leaving agent milestones to the firm.

Sources

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