Book a demo
Topics · GOVERNANCE-AGENTS

Asset management AI agent registry Risks | Moring

The risks of running asset management agents without a registry include invisible agent sprawl, unclear accountability, permission creep, untracked model dependencies, and no evidence trail for audit. A registry addresses each risk by making every agent a named, owned, and reviewable record with defined scope and controls.

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

Short answer

The risks of running asset management agents without a registry include invisible agent sprawl, unclear accountability, permission creep, untracked model dependencies, and no evidence trail for audit. A registry addresses each risk by making every agent a named, owned, and reviewable record with defined scope and controls.

How it works

The core risk of an ungoverned agent estate is that the estate becomes unknowable. Agents are easy to build: the Cloud Security Alliance's draft Agent Identity Governance Framework, published in March 2026, observes that AI agents are a class of identity that does not fit the assumptions of traditional identity and access management. Without a registry, an asset manager accumulates autonomous actors that hold credentials, reach data, and take actions, but that appear in no inventory and answer to no owner.

Each unregistered agent compounds specific risks. Accountability risk: when something goes wrong in a research summary, a client report, or a trade support action, nobody is identified as responsible. Permission risk: the agent inherits broad credentials from its developer or runtime, exceeding any approved scope. Dependency risk: the agent consumes model outputs that were validated for a different context, breaking the traceability that SR 11-7 style model risk management expects. Evidence risk: there is no record of what the agent did, so audit and EU AI Act obligations cannot be met with confidence.

The Cloud Security Alliance framework also identifies non-human identity governance gaps that map directly onto agents: identities created outside formal process, credentials that never expire, and access that nobody reviews. Palo Alto's agentic AI governance guide makes the same point from the security side, describing governance as setting boundaries on what agents can access and perform at runtime. A registry is the inventory layer that makes both sets of boundaries enforceable.

Sprawl has a measurable dimension as well. Deloitte's 2026 State of AI in the Enterprise survey of 3,235 leaders across 24 countries found only one in five organizations has a mature model for governing autonomous agents, which implies most estates are growing faster than their governance. In asset management, where agents touch fiduciary data, the gap between deployment speed and governance capacity is itself a risk that examiners and boards will ask about.

A registry converts these risks into manageable ones. Registration forces accountability at creation; recorded permissions cap scope; recorded model and tool dependencies restore traceability; and the registry's review cycle catches drift before it becomes an audit finding. None of this eliminates agent risk, but it changes the risk from unknown to known, which is the precondition for managing it under the model risk discipline asset managers already apply.

The residual lesson is sequencing. Firms that build the registry early treat agents as governed assets from day one; firms that add it later face an inventory archaeology project, reconstructing ownership and scope for agents already in production. The Cloud Security Alliance's framework framing is useful here: govern agent identities as deliberately as human ones, and the registry is where that governance becomes concrete.

Frequently asked questions

What risks arise when asset managers deploy agents without a registry?

The principal risks are invisibility and accountability gaps: agents hold credentials, reach data, and take actions but appear in no inventory and answer to no owner. This creates permission creep beyond approved scope, untracked dependencies on models that may not be validated for that context, and no evidence trail for audit. The Cloud Security Alliance's 2026 draft framework on agent identity governance highlights exactly this mismatch with traditional access management, and in asset management the exposure lands on fiduciary data.

How does a registry reduce agent accountability risk?

A registry reduces accountability risk by requiring a named owner and a documented purpose at registration, before an agent runs in production. When an agent produces a flawed research summary or acts incorrectly in trade support, the record identifies who is responsible and what scope the agent held. This mirrors SR 11-7 model risk management, where every model has an accountable owner; extending it to agents closes the gap Deloitte's 2026 survey found, with only one firm in five governing agents at a mature level.

How does the registry address non-human identity risks for agents?

The Cloud Security Alliance's draft Agent Identity Governance Framework identifies gaps in non-human identity management: identities created outside formal process, credentials that never expire, and access that nobody reviews. A registry addresses these by making registration the formal path through which an agent identity is created, recording its permissions and tools, and subjecting both to recurring review. In an asset management firm this ensures agent identities touching portfolio or client data are governed as deliberately as human access.

Why is agent sprawl a regulatory concern for asset managers?

Sprawl is a regulatory concern because undocumented autonomous systems cannot demonstrate the oversight that regulators expect. The EU AI Act imposes documentation and risk classification duties on deployers of AI systems, and SR 11-7 style model risk management expects inventories and monitoring of models in use. An ungoverned agent estate defeats both: nothing documents what runs, and nothing monitors what it does. A registry restores the inventory and monitoring, converting sprawl from an unmanaged exposure into a governed population of agents.

Sources

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