Question 570
At fictional Amberlight Energy, an acceptance review of an AI-assisted code review finds: A conversational assistant intended for route lookup is accidentally granted Terraform apply permissions to all campus workspaces. Which engineering principle explains the described system behavior?
Explanation
Correct Answer
The model is not a reliable security principal; tool identities and server-side authorization bound the damage of agent errors. The corresponding repair is: Issue read-only scoped identity credentials for diagnostic agents and place separate approval boundaries around mutating tools. Acceptance should show: Authorization tests show route queries succeed but unapproved configuration changes are rejected before reaching Cisco APIs. Applied here, the relevant facts are: A conversational assistant intended for route lookup is accidentally granted Terraform apply permissions to all campus workspaces. The question asks which engineering principle explains the described system behavior.
Incorrect Answers
The offered statement is true for this other mechanism: The corresponding repair is: Evaluate the agent on a representative labeled test set with difficult, adversarial and negative cases, measuring factual error and unsafe-action rates. It does not explain the facts described here, which follow from: The corresponding repair is: Issue read-only scoped identity credentials for diagnostic agents and place separate approval boundaries around mutating tools.
The corresponding repair is: Separate read-only diagnostic tools from mutating tools and require an explicit scoped approval step before any privileged operation. That explanation applies to a different technical failure. Here the controlling principle is: The corresponding repair is: Issue read-only scoped identity credentials for diagnostic agents and place separate approval boundaries around mutating tools.
The offered statement is true for this other mechanism: The corresponding repair is: Use AI assistance for bounded drafts and test generation while assigning accountable reviewers and validating facts against approved sources. It does not explain the facts described here, which follow from: The corresponding repair is: Issue read-only scoped identity credentials for diagnostic agents and place separate approval boundaries around mutating tools.