100% Latest & Updated Cisco 350-901 Practice Test Questions, Exam Dumps & Verified Answers!
30 Days Free Updates, Instant Download!
Question 920
The agent uses yesterday's cached device inventory when responding to 'which devices are down now?'. Independently, reviewers flag this potentially unsafe practice for assessment: 'Merge all old incidents into one current outage report automatically'. Which verification addresses that practice without assuming it caused the incident?
- Fill missing sites with assumed healthy values
- Inspect event and ticket linkage
- Inspect fresh tool result
- Inspect live telemetry query time
Explanation
Correct Answer
B:Unverified correlation exaggerates scope. The intended behavior is: Context improves interpretation when bounded.
Incorrect Answers
A:Fabrication masks monitoring gaps.
C:This checks query authoritative state, not the named failure. Current evidence is required for current status.
D:This checks separate memory from live data, not the named failure. Memory can become stale.
Question 921
An operations agent is allowed to propose network maintenance commands but not execute them directly. The specific implementation objective is to produce a reviewable plan. Which configuration best meets the stated requirement?
- Allow the model to override change restrictions within its own text
- Generate intended devices, commands, diffs and rollback as a proposal
- Mark operation complete from the model's intended outcome description
- Interpret silence as approval for a dangerous change
Explanation
Correct Answer
B:Plans can be reviewed before side effects.
Incorrect Answers
A:Self-authorization is not enforcement.
C:Intention does not prove state.
D:No response is not authorization.
Question 922
An operations agent is allowed to propose network maintenance commands but not execute them directly. The specific implementation objective is to apply independent change policy. Which design action is most appropriate for this case?
- Send proposed commands straight to devices as soon as they are generated
- Interpret silence as approval for a dangerous change
- Validate generated plan against approved access and blast-radius limits
- Mark operation complete from the model's intended outcome description
Explanation
Correct Answer
C:Policy checks should not depend solely on model compliance.
Incorrect Answers
A:Unreviewed execution may break service.
B:No response is not authorization.
D:Intention does not prove state.
Question 923
An operations agent is allowed to propose network maintenance commands but not execute them directly. The specific implementation objective is to require human decision. What should the engineer implement for this specific objective?
- Obtain explicit approver input for privileged actions
- Allow the model to override change restrictions within its own text
- Mark operation complete from the model's intended outcome description
- Send proposed commands straight to devices as soon as they are generated
Explanation
Correct Answer
A:Human approval separates suggestion from execution.
Incorrect Answers
B:Self-authorization is not enforcement.
C:Intention does not prove state.
D:Unreviewed execution may break service.
Question 924
An operations agent is allowed to propose network maintenance commands but not execute them directly. The specific implementation objective is to verify with real tests. Which adjustment would satisfy this particular constraint?
- Send proposed commands straight to devices as soon as they are generated
- Allow the model to override change restrictions within its own text
- After approval and execution, validate desired and denied paths on targeted systems
- Interpret silence as approval for a dangerous change
Explanation
Correct Answer
C:Successful tool invocation is not service correctness.
Incorrect Answers
A:Unreviewed execution may break service.
B:Self-authorization is not enforcement.
D:No response is not authorization.
Question 925
An operations agent is allowed to propose network maintenance commands but not execute them directly. The deployment plan must both (1) produce a reviewable plan and (2) require human decision. Which TWO actions together satisfy those requirements? Choose TWO.
- Obtain explicit approver input for privileged actions
- Send proposed commands straight to devices as soon as they are generated
- Generate intended devices, commands, diffs and rollback as a proposal
- Interpret silence as approval for a dangerous change
- Allow the model to override change restrictions within its own text
Explanation
Correct Answers
A:Human approval separates suggestion from execution.
C:Plans can be reviewed before side effects.
Incorrect Answers
B:Unreviewed execution may break service.
D:No response is not authorization.
E:Self-authorization is not enforcement.
Question 926
An operations agent is allowed to propose network maintenance commands but not execute them directly. Three conditions must be met: (1) apply independent change policy; (2) require human decision; and (3) verify with real tests. Which THREE actions form the narrowest complete plan? Choose THREE.
- Mark operation complete from the model's intended outcome description
- After approval and execution, validate desired and denied paths on targeted systems
- Obtain explicit approver input for privileged actions
- Allow the model to override change restrictions within its own text
- Validate generated plan against approved access and blast-radius limits
- Send proposed commands straight to devices as soon as they are generated
Explanation
Correct Answers
B:Successful tool invocation is not service correctness.
C:Human approval separates suggestion from execution.
E:Policy checks should not depend solely on model compliance.
Incorrect Answers
A:Intention does not prove state.
D:Self-authorization is not enforcement.
F:Unreviewed execution may break service.
Question 927
A prompt injection in a retrieved runbook tries to make the agent run a global ACL removal. Independently, reviewers flag this potentially unsafe practice for assessment: 'Send proposed commands straight to devices as soon as they are generated'. Which inspection most directly evaluates that practice?
- Inspect approval evidence
- Allow the model to override change restrictions within its own text
- Inspect readback and negative test results
- Inspect plan artifact and tool permissions
Explanation
Correct Answer
D:Unreviewed execution may break service. The intended behavior is: Plans can be reviewed before side effects.
Incorrect Answers
A:This checks require human decision, not the named failure. Human approval separates suggestion from execution.
B:Self-authorization is not enforcement.
C:This checks verify with real tests, not the named failure. Successful tool invocation is not service correctness.
Question 928
A prompt injection in a retrieved runbook tries to make the agent run a global ACL removal. Independently, reviewers flag this potentially unsafe practice for assessment: 'Allow the model to override change restrictions within its own text'. What diagnostic evidence should first be collected to assess that practice?
- Inspect plan artifact and tool permissions
- Inspect external policy-gate result
- Mark operation complete from the model's intended outcome description
- Inspect approval evidence
Explanation
Correct Answer
B:Self-authorization is not enforcement. The intended behavior is: Policy checks should not depend solely on model compliance.
Incorrect Answers
A:This checks produce a reviewable plan, not the named failure. Plans can be reviewed before side effects.
C:Intention does not prove state.
D:This checks require human decision, not the named failure. Human approval separates suggestion from execution.
Question 929
A prompt injection in a retrieved runbook tries to make the agent run a global ACL removal. Independently, reviewers flag this potentially unsafe practice for assessment: 'Interpret silence as approval for a dangerous change'. Which focused check best tests this specific assumption?
- Inspect plan artifact and tool permissions
- Allow the model to override change restrictions within its own text
- Inspect readback and negative test results
- Inspect approval evidence
Explanation
Correct Answer
D:No response is not authorization. The intended behavior is: Human approval separates suggestion from execution.
Incorrect Answers
A:This checks produce a reviewable plan, not the named failure. Plans can be reviewed before side effects.
B:Self-authorization is not enforcement.
C:This checks verify with real tests, not the named failure. Successful tool invocation is not service correctness.
Question 930
A prompt injection in a retrieved runbook tries to make the agent run a global ACL removal. Independently, reviewers flag this potentially unsafe practice for assessment: 'Mark operation complete from the model's intended outcome description'. Which verification addresses that practice without assuming it caused the incident?
- Inspect external policy-gate result
- Inspect readback and negative test results
- Interpret silence as approval for a dangerous change
- Inspect plan artifact and tool permissions
Explanation
Correct Answer
B:Intention does not prove state. The intended behavior is: Successful tool invocation is not service correctness.
Incorrect Answers
A:This checks apply independent change policy, not the named failure. Policy checks should not depend solely on model compliance.
C:No response is not authorization.
D:This checks produce a reviewable plan, not the named failure. Plans can be reviewed before side effects.
Designing, Deploying, and Managing Network Automation Systems (350-901) Practice Tests