Question 450
At fictional Sunrise Aeronautics, during an AI-assisted code review, the operator observes: An AI chat assistant translates free-text "fix routing" into unbounded commands rather than requiring a precise change proposal. What acceptance check would prove the intended behavior was restored?
Explanation
Correct Answer
The assistant outputs a reviewable structured plan and no device mutation occurs until the scoped proposal is approved. This directly verifies the related correction: Convert user requests into a typed proposed intent with explicit targets, preconditions and a separate human-approved execution path. It checks the technical distinction: LLMs are useful for language interpretation but the execution boundary should accept validated structured intents, not arbitrary prose. Applied here, the relevant facts are: An AI chat assistant translates free-text "fix routing" into unbounded commands rather than requiring a precise change proposal. The question asks what acceptance check would prove the intended behavior was restored.
Incorrect Answers
That test belongs to another control with this basis: This directly verifies the related correction: Use AI assistance for bounded drafts and test generation while assigning accountable reviewers and validating facts against approved sources. In the current case, the governing reason is: This directly verifies the related correction: Convert user requests into a typed proposed intent with explicit targets, preconditions and a separate human-approved execution path. The direct verification is “The assistant outputs a reviewable structured plan and no device mutation occurs until the scoped proposal is approved”.
The alternative verifies a different condition: This directly verifies the related correction: Evaluate the agent on a representative labeled test set with difficult, adversarial and negative cases, measuring factual error and unsafe-action rates. The current case instead depends on: This directly verifies the related correction: Convert user requests into a typed proposed intent with explicit targets, preconditions and a separate human-approved execution path. Confirm that “The assistant outputs a reviewable structured plan and no device mutation occurs until the scoped proposal is approved”.
That test belongs to another control with this basis: This directly verifies the related correction: Define typed, validated tool arguments with explicit device allowlists and reject invalid or unauthorized inputs at the server boundary. In the current case, the governing reason is: This directly verifies the related correction: Convert user requests into a typed proposed intent with explicit targets, preconditions and a separate human-approved execution path. The direct verification is “The assistant outputs a reviewable structured plan and no device mutation occurs until the scoped proposal is approved”.