Question 430
At fictional Amberlight Energy, an acceptance review of an AI-assisted code review finds: An incident chatbot interprets a troubleshooting question as authority to restart an OSPF process on production edge routers. Which sustainable preventive control most directly avoids a recurrence?
Explanation
Correct Answer
Implement tool categorization, approval records and run-as identities for every privileged network agent operation. This prevents the recurrence because agent tool chaining can enlarge the blast radius of an ambiguous user instruction unless execution permissions are independently gated. A matching immediate correction is: Separate read-only diagnostic tools from mutating tools and require an explicit scoped approval step before any privileged operation. Applied here, the relevant facts are: An incident chatbot interprets a troubleshooting question as authority to restart an OSPF process on production edge routers. The question asks which sustainable preventive control most directly avoids a recurrence.
Incorrect Answers
This routine guards against another mechanism: tool wrappers must preserve the meaning of missing, stale and failed data for the consumer agent. The recurring exposure in the current case is: agent tool chaining can enlarge the blast radius of an ambiguous user instruction unless execution permissions are independently gated. The matching preventive check is “Implement tool categorization, approval records and run-as identities for every privileged network agent operation”.
The suggested review tracks a different technical risk: aI recommendations are hypotheses requiring external ground truth; absent evidence is not proof of a network condition. Here the sustained control must address: agent tool chaining can enlarge the blast radius of an ambiguous user instruction unless execution permissions are independently gated. Choose “Implement tool categorization, approval records and run-as identities for every privileged network agent operation”.
That safeguard is relevant to: retrieval-augmented systems can inherit false or hostile data; source quality and technical cross-checks must precede action. The current failure instead follows from: agent tool chaining can enlarge the blast radius of an ambiguous user instruction unless execution permissions are independently gated. Its prevention measure is “Implement tool categorization, approval records and run-as identities for every privileged network agent operation”.