AB-100 practice question 31 of 35

After a prompt update, your Copilot Studio agent begins returning internal case notes that should be restricted to supervisors, and the incident is…

Full set: 708 questions
After a prompt update, your Copilot Studio agent begins returning internal case notes that should be restricted to supervisors, and the incident is escalated under your regulated- data policy. You must remediate within one hour without redeploying the consuming apps, and you need an audit trail that ties the mitigation to a specific prompt version. The issue appears only in production where the new prompt package was promoted last night. Requirements • Restore compliant behavior within one hour • Preserve versioned audit evidence for investigators • Avoid app redeploy or connector changes Which action best resolves the incident while validating prompt best practices going forward?
Answer options
A: Disable all connectors until security review completes
B: Raise model temperature to reduce memorization
C: Rollback prompt library version Correct answer
D: Add tenant-wide sensitivity label rule
Correct Answer:
C. Rollback prompt library version
Explanation

Rolling back the prompt library to the previous, stable version is the most direct and fastest way to remediate the data leakage incident. This action specifically targets the root cause—the faulty prompt update—and restores compliant behavior within the one-hour SLA without requiring application redeployment. Using a version-controlled system for the prompt library ensures that this rollback is a simple, auditable action, preserving a clear trail for investigators by linking the incident to a specific version change.

Why Incorrect

A. Disabling all connectors is an overly broad action that would cause a complete service outage, violating the principle of a targeted fix.
B. Raising model temperature is irrelevant to fixing a prompt-induced data leakage issue and could make outputs less predictable.
D. Adding a sensitivity label is a long-term governance control, not an immediate incident response action that can be completed within one hour.

References

1. Microsoft

"Responsible AI Standard

v2

" (June 2022). Section 4.3

"Reliability and Safety

" requires mechanisms to "prevent and mitigate potential harms

" for which a rapid rollback capability is a primary control.

2. Microsoft Azure Well-Architected Framework

"Design for operations." This principle advocates for designing systems that are easy to manage and can be quickly restored to a known-good state in case of failure

with versioning and rollback being key tactics.

3. Microsoft Azure AI Studio Documentation

"Source control for prompt flow." Integrating prompt development with source control (like Git) enables versioning

which is a prerequisite for effective and auditable rollback procedures.

Scroll to Top