Question 30
While investigating an AI recommendation accuracy study for fictional Sunrise Aeronautics, the team reports: A conversational network assistant is permitted to run a harmless ping but then automatically chains into an unapproved ACL modification. What acceptance check would prove the intended behavior was restored?
Explanation
Correct Answer
A conversational diagnostic completes safely without write access, and a proposed write remains pending until an authorized approver confirms. This directly verifies the related correction: Separate read-only diagnostic tools from mutating tools and require an explicit scoped approval step before any privileged operation. It checks the technical distinction: Agent tool chaining can enlarge the blast radius of an ambiguous user instruction unless execution permissions are independently gated.
Incorrect Answers
That test belongs to another control with this basis: This directly verifies the related correction: Sanitize inputs and use an approved private or contractually controlled AI service without exposing secrets or regulated network data. In the current case, the governing reason is: This directly verifies the related correction: Separate read-only diagnostic tools from mutating tools and require an explicit scoped approval step before any privileged operation. The direct verification is “A conversational diagnostic completes safely without write access, and a proposed write remains pending until an authorized approver confirms”.
This observation would test a different technical dependency: This directly verifies the related correction: Restrict tool egress to approved destinations and never grant untrusted retrieved text authority to choose exfiltration endpoints. The current issue is governed by: This directly verifies the related correction: Separate read-only diagnostic tools from mutating tools and require an explicit scoped approval step before any privileged operation. Its decisive check is “A conversational diagnostic completes safely without write access, and a proposed write remains pending until an authorized approver confirms”.
That test belongs to another control with this basis: This directly verifies the related correction: Ground actionable recommendations in trusted versioned technical references and reject unsupported or conflicting retrieved claims. In the current case, the governing reason is: This directly verifies the related correction: Separate read-only diagnostic tools from mutating tools and require an explicit scoped approval step before any privileged operation. The direct verification is “A conversational diagnostic completes safely without write access, and a proposed write remains pending until an authorized approver confirms”.