CCAR-P Governance, Safety, and Risk Management Practice Question

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✗

  Route all agent traffic through a proxy that logs every tool invocation and alerts the security operations center after transfers complete.

  Why it's wrong here

  Detection after the fact does not prevent an unauthorized transfer from succeeding. Logging and alerting are valuable for investigation and response, but the funds have already moved by the time an alert fires. Because the requirement is to limit the blast radius of unauthorized transfers, a purely detective control leaves the primary harm unmitigated and is insufficient on its own.

• ✓

  Enforce authorization and transaction limits in the downstream banking APIs the agent calls, independent of anything the model outputs.

  Why this is correct

  This is correct because placing authorization, per-transaction caps, and velocity limits in the downstream systems means the model's output can never exceed what the API permits, regardless of manipulation. The blast radius is bounded by deterministic server-side policy rather than by model behavior, which is exactly the defense-in-depth posture risk leadership is requesting for a high-impact tool.

• ✗

  Add a system prompt instructing the model to never perform transfers above a defined threshold or to accounts not previously seen.

  Why it's wrong here

  Prompt-level instructions are advisory, not enforcing. A manipulated or jailbroken agent can be steered around textual constraints, and the model has no reliable way to guarantee compliance under adversarial input. Because the requirement is to bound the blast radius of unauthorized transfers, a control that lives only in the prompt provides no hard guarantee and fails the risk objective.

• ✗

  Increase the model's temperature to zero so its responses become fully deterministic and cannot be manipulated.

  Why it's wrong here

  Temperature controls sampling randomness, not adversarial robustness. A deterministic model can still be induced to emit a harmful tool call if the input is crafted to steer it, and determinism says nothing about whether that call is authorized. Lowering temperature does not create any enforcement boundary around fund transfers, so it does not reduce the blast radius of a manipulated agent.