Asset Mgmt Agent Governance Access Control | Moring
Access control is the load-bearing element of AI agent governance for asset managers: agents touch portfolios, client data and reporting systems under identities whose scopes determine institutional risk. The design pattern is unique, task-scoped identities for every agent, least privilege across data and systems, and access decisions logged as evidence. SR 26-2 and SR 11-7 supervisory expectations translate directly to the agent layer.
Short answer
Access control is the load-bearing element of AI agent governance for asset managers: agents touch portfolios, client data and reporting systems under identities whose scopes determine institutional risk. The design pattern is unique, task-scoped identities for every agent, least privilege across data and systems, and access decisions logged as evidence. SR 26-2 and SR 11-7 supervisory expectations translate directly to the agent layer.
How it works
Asset managers adopting agentic AI inherit a control question older than the technology: who may touch what, under which identity, with what evidence. Agents execute work across research, fund accounting, client reporting and compliance workflows at a velocity human processes were never designed to gate, and access control is the governance element that determines whether that velocity happens inside institutional rules or outside them. Dataiku's governance guide lists access and identity controls among the five components of a production governance program, and for a regulated firm they are the component examiners test first.
The baseline is identity discipline. Each agent should hold a unique, traceable identity, never shared API keys, with full logging of its actions. This matters doubly in asset management because attribution is what SR 26-2 and SR 11-7 model risk expectations depend on: when an agent-generated report or reconciliation reaches a client or a regulator, the firm must be able to say which system produced it, under whose authority and within what scope. A shared credential makes that reconstruction impossible and converts a technical shortcut into an accountability failure.
Least privilege is the second discipline. Airlock Digital's governance guidance restricts agents to only the systems, data and actions required for assigned tasks, and PwC's workforce governance research states the goal plainly: each agent should have a verified identity, defined role, task-specific permissions and auditable records, with access governance reflecting industry risk, organizational role and task scope. An agent drafting client letters needs read access to approved holdings data, not trading authority; an agent supporting NAV reconciliation needs the accounting ledger, not the order management system. Scope creep is how a reporting assistant becomes an unexamined trading participant.
Delegated authority is the third discipline. Zenity's CISO guidance observes that agents inherit delegated identities and API access and scale faster than governance programs, which is why scope must be provisioned deliberately rather than accumulated. Sweet's governance guidance checks whether each delegated action is expected, permitted and reversible before an agent changes systems or data, and Microsoft's cloud adoption framework recommends a centralized, enforceable baseline for all AI agents aligned with existing identity, data governance and security practices. The pattern is an extension of how the firm already governs human access, not a parallel regime invented for agents.
Evidence is the fourth discipline. Access decisions, grants, revocations, scope changes and rejections should be logged and joined to the audit trail, because the EU AI Act's traceability and documentation expectations for high-risk systems assume the firm can reconstruct who did what and under what permission. Akamai's guidance identifies shadow AI discovery as a critical first step, since enforcement has blind spots without a complete inventory of authorized and unauthorized agents. The completed access-control layer is then more than a security control: it is the mechanism that makes agent adoption in asset management defensible to boards, clients and examiners alike.
Sources:
Frequently asked questions
Why are unique agent identities a supervisory issue rather than just a security practice?
SR 26-2 and SR 11-7 model risk expectations require asset managers to reconstruct which system produced a report, reconciliation or client communication, under whose authority and within what scope. Shared API keys make that reconstruction impossible, converting a technical shortcut into an accountability failure. Dataiku's governance guide lists access and identity controls among the five components of a production governance program, and examiners test this component first because attribution underpins every other control they review.
How should least privilege be applied to an asset management agent fleet?
Airlock Digital's guidance restricts agents to only the systems, data and actions required for their assigned tasks, and PwC's workforce governance research specifies a verified identity, defined role, task-specific permissions and auditable records per agent. In practice, an agent drafting client letters needs read access to approved holdings data but not trading authority, while a NAV reconciliation agent needs the accounting ledger rather than the order management system. Scope creep is how a reporting assistant becomes an unexamined trading participant.
What does delegated authority mean for agents in a regulated asset manager?
Zenity's guidance observes that agents inherit delegated identities and API access and scale faster than governance programs, so scope must be provisioned deliberately rather than accumulated. Sweet's governance guidance checks whether each delegated action is expected, permitted and reversible before an agent changes systems or data. Microsoft's cloud adoption framework recommends a centralized, enforceable baseline aligned with existing identity and data governance practices, meaning agent access extends the firm's existing human-access regime rather than creating a parallel one.
Why must access decisions be logged and joined to the audit trail?
The EU AI Act's traceability and documentation expectations for high-risk systems assume the firm can reconstruct who did what under which permission, so grants, revocations, scope changes and rejections must be recorded with the audit trail. Akamai adds that shadow AI discovery is critical because enforcement has blind spots without a complete inventory of authorized and unauthorized agents. Together these records make agent adoption defensible to boards, clients and examiners rather than merely restricting technical access.