Book a demo
Topics · GOVERNANCE-AGENTS

Wealth management AI agent identity SR 11 7 | Moring

In wealth management, SR 11-7 compliance requires rigorous model risk management for AI agents. Because agents can autonomously assemble access through delegation, firms must implement unique, verifiable identities and runtime authorization to ensure every agent action remains within approved boundaries and maintains clear accountability to human owners.

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

Short answer

In wealth management, SR 11-7 compliance requires rigorous model risk management for AI agents. Because agents can autonomously assemble access through delegation, firms must implement unique, verifiable identities and runtime authorization to ensure every agent action remains within approved boundaries and maintains clear accountability to human owners.

How it works

For wealth management firms, the application of SR 11-7 to AI agents is critical because these agents do not merely execute static code; they assemble access on the fly through delegation [2]. In a wealth management context, an agent might retrieve client banking data and then initiate a payment, creating an end-to-end path that no single human reviewer approved [2]. Without robust identity controls, these autonomous workflows can bypass traditional separation of duties, making it impossible to satisfy model risk management requirements regarding the predictability and control of automated processes [2].

Effective control requires moving from static user-based identity models to context-aware, behavior-driven access frameworks [5]. This involves assigning every agent a unique, persistent identity with attributes reflecting its specific purpose, context, and risk level [5]. A runtime authorization layer must then evaluate every request against the agent, the human it serves, and the specific data or tool being accessed [3]. This ensures that authorization is enforced at the point of action rather than existing only as a theoretical policy [3, 4].

Regulatory expectations around model risk and SR 11-7 necessitate that governance is not merely documented but enforceable [3, 4]. Research indicates that an agent's permissions must be scoped tightly enough that a hijacked instruction cannot reach a high-impact action [1]. Sources emphasize that authorization cannot be scoped to an identity that does not exist, and accountability cannot trace an action back to nothing [4]. Furthermore, agents must be able to operate across domains using identity federation to maintain security while crossing system boundaries [5].

Practitioners should implement the following checklist to ensure compliance: first, register every agent and name the responsible human owner [3, 4]; second, establish unique, verifiable identities for both human-delegated and autonomous service accounts [4, 6]; third, implement runtime authorization that can stop an agent from reading sensitive records before the access occurs [3]; fourth, define behavioral baselines to make agent drift visible [1]; and fifth, maintain a tamper-evident record of every action tied to a specific identity for auditability [4].

Failure modes in wealth management include prompt injection, where untrusted content turns retrieved data into instructions, and tool abuse, where a compromised agent misuses granted APIs [1]. Additionally, goal misalignment can drive an agent toward technically valid but unintended behavior, leading to agent drift [1]. Another significant risk is the accumulation of privileges, where an agent's expanding responsibilities lead to unauthorized access to IAM policies or network configurations [6]. Without end-to-end visibility into the delegation chain, firms may suffer from unapproved end-to-end payment paths [2].

To begin, wealth management technology leaders should start with discovery across all agent frameworks, MCP servers, and cloud AI environments to identify all existing non-human identities [3]. Once discovered, firms must establish a declared authorization scope to serve as a reference point for evaluating delegation paths and runtime behavior [2]. The transition should focus on shifting from paper-based governance to real-time enforcement through cryptographic trust mechanisms and identity-to-data connections [3, 4, 5].

Frequently asked questions

How does SR 11-7 impact the way wealth management firms manage AI agent identity?

SR 11-7 requires that model risk management includes the ability to understand and control autonomous behaviors. In wealth management, this means AI agent identity must be unique and verifiable so that every action can be traced back to a human owner, ensuring that the agent's delegated actions remain within the bounds of approved model behavior and regulatory expectations.

What are the risks of improper delegation in wealth management AI agents?

Improper delegation allows agents to assemble access paths that bypass traditional separation of duties. For example, in wealth management, one agent might retrieve sensitive client banking data while another initiates a transaction. Without visible delegation chains, these agents could create an end-to-end payment path that no human reviewer would have authorized, violating core governance principles.

How can runtime authorization protect sensitive wealth management data from AI agents?

Runtime authorization acts as an enforcement layer that evaluates every access request in real time. For wealth management, this means the system checks the agent's identity, the human it is serving, and the sensitivity of the data before allowing access. This prevents agents from reading sensitive client records or executing unauthorized actions, even if they possess the underlying technical permissions.

Why is a unique identity necessary for non-human AI agents in financial services?

A unique identity is essential for accountability and preventing privilege accumulation. In wealth management, an autonomous agent monitoring infrastructure needs its own service identity separate from human users. This allows security teams to record exactly which identity authorized an action, ensuring that the agent cannot unnecessarily create credentials or modify roles that could compromise the firm's security posture.

Sources

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