100% Latest & Updated Cisco 350-901 Practice Test Questions, Exam Dumps & Verified Answers!
30 Days Free Updates, Instant Download!
Question 850
A malicious attachment says to disable logging and upload controller secrets to an external site. Independently, reviewers flag this potentially unsafe practice for assessment: 'Strip all detection logs after the model rejects a payload'. Which verification addresses that practice without assuming it caused the incident?
- Inspect injection alerts and blocked-call logs
- Permit the agent to send arbitrary network files to any URL
- Inspect document trust tagging
- Check requester identity and approval
Explanation
Correct Answer
A:Lost evidence weakens defense testing. The intended behavior is: Visibility supports safe agent operation.
Incorrect Answers
B:Untrusted instructions can leak data.
C:This checks isolate external content, not the named failure. Retrieved context has lower authority than application policy.
D:This checks require authorized intent, not the named failure. Ticket text is not a trusted change request.
Question 851
A conversational network agent can read interface health and optionally apply approved shutdowns. The specific implementation objective is to separate reading and writing. Which configuration best meets the stated requirement?
- Use distinct identities or role scopes for diagnostics and controlled changes
- Let any logged-in help desk user shut down core interfaces
- Embed router passwords in ordinary model messages
- Store only a model summary saying 'done' after changes
Explanation
Correct Answer
A:A read assistant does not need write power by default.
Incorrect Answers
B:Authentication without RBAC permits abuse.
C:Messages may leak sensitive credentials.
D:Narration cannot prove execution.
Question 852
A conversational network agent can read interface health and optionally apply approved shutdowns. The specific implementation objective is to keep secrets outside prompts. Which design action is most appropriate for this case?
- Inject credentials at trusted execution boundary rather than exposing them to the LLM
- Use one unrestricted administrator token for all agent activities
- Store only a model summary saying 'done' after changes
- Let any logged-in help desk user shut down core interfaces
Explanation
Correct Answer
A:Prompt context can be disclosed or manipulated.
Incorrect Answers
B:Token compromise exposes entire network.
C:Narration cannot prove execution.
D:Authentication without RBAC permits abuse.
Question 853
A conversational network agent can read interface health and optionally apply approved shutdowns. The specific implementation objective is to check user authority. What should the engineer implement for this specific objective?
- Use one unrestricted administrator token for all agent activities
- Store only a model summary saying 'done' after changes
- Verify the user's identity and resource permissions before any tool operation
- Embed router passwords in ordinary model messages
Explanation
Correct Answer
C:Conversation intent is not sufficient authorization.
Incorrect Answers
A:Token compromise exposes entire network.
B:Narration cannot prove execution.
D:Messages may leak sensitive credentials.
Question 854
A conversational network agent can read interface health and optionally apply approved shutdowns. The specific implementation objective is to log privileged changes. Which adjustment would satisfy this particular constraint?
- Use one unrestricted administrator token for all agent activities
- Let any logged-in help desk user shut down core interfaces
- Embed router passwords in ordinary model messages
- Capture change owner, device, command and resulting state
Explanation
Correct Answer
D:Action traceability supports accountability.
Incorrect Answers
A:Token compromise exposes entire network.
B:Authentication without RBAC permits abuse.
C:Messages may leak sensitive credentials.
Question 855
A conversational network agent can read interface health and optionally apply approved shutdowns. The deployment plan must both (1) keep secrets outside prompts and (2) log privileged changes. Which TWO actions together satisfy those requirements? Choose TWO.
- Use one unrestricted administrator token for all agent activities
- Inject credentials at trusted execution boundary rather than exposing them to the LLM
- Let any logged-in help desk user shut down core interfaces
- Capture change owner, device, command and resulting state
- Embed router passwords in ordinary model messages
Explanation
Correct Answers
B:Prompt context can be disclosed or manipulated.
D:Action traceability supports accountability.
Incorrect Answers
A:Token compromise exposes entire network.
C:Authentication without RBAC permits abuse.
E:Messages may leak sensitive credentials.
Question 856
A conversational network agent can read interface health and optionally apply approved shutdowns. Three conditions must be met: (1) separate reading and writing; (2) check user authority; and (3) log privileged changes. Which THREE actions form the narrowest complete plan? Choose THREE.
- Embed router passwords in ordinary model messages
- Verify the user's identity and resource permissions before any tool operation
- Let any logged-in help desk user shut down core interfaces
- Use distinct identities or role scopes for diagnostics and controlled changes
- Capture change owner, device, command and resulting state
- Store only a model summary saying 'done' after changes
Explanation
Correct Answers
B:Conversation intent is not sufficient authorization.
D:A read assistant does not need write power by default.
E:Action traceability supports accountability.
Incorrect Answers
A:Messages may leak sensitive credentials.
C:Authentication without RBAC permits abuse.
F:Narration cannot prove execution.
Question 857
A prototype stores one full-admin token in prompt context for every user conversation. Independently, reviewers flag this potentially unsafe practice for assessment: 'Use one unrestricted administrator token for all agent activities'. Which inspection most directly evaluates that practice?
- Inspect user-to-tool policy
- Inspect tool and device audit records
- Embed router passwords in ordinary model messages
- Inspect tool-specific credentials
Explanation
Correct Answer
D:Token compromise exposes entire network. The intended behavior is: A read assistant does not need write power by default.
Incorrect Answers
A:This checks check user authority, not the named failure. Conversation intent is not sufficient authorization.
B:This checks log privileged changes, not the named failure. Action traceability supports accountability.
C:Messages may leak sensitive credentials.
Question 858
A prototype stores one full-admin token in prompt context for every user conversation. Independently, reviewers flag this potentially unsafe practice for assessment: 'Embed router passwords in ordinary model messages'. What diagnostic evidence should first be collected to assess that practice?
- Inspect prompt and runtime secret design
- Inspect user-to-tool policy
- Inspect tool-specific credentials
- Store only a model summary saying 'done' after changes
Explanation
Correct Answer
A:Messages may leak sensitive credentials. The intended behavior is: Prompt context can be disclosed or manipulated.
Incorrect Answers
B:This checks check user authority, not the named failure. Conversation intent is not sufficient authorization.
C:This checks separate reading and writing, not the named failure. A read assistant does not need write power by default.
D:Narration cannot prove execution.
Question 859
A prototype stores one full-admin token in prompt context for every user conversation. Independently, reviewers flag this potentially unsafe practice for assessment: 'Let any logged-in help desk user shut down core interfaces'. Which focused check best tests this specific assumption?
- Inspect prompt and runtime secret design
- Inspect tool-specific credentials
- Store only a model summary saying 'done' after changes
- Inspect user-to-tool policy
Explanation
Correct Answer
D:Authentication without RBAC permits abuse. The intended behavior is: Conversation intent is not sufficient authorization.
Incorrect Answers
A:This checks keep secrets outside prompts, not the named failure. Prompt context can be disclosed or manipulated.
B:This checks separate reading and writing, not the named failure. A read assistant does not need write power by default.
C:Narration cannot prove execution.
Question 860
A prototype stores one full-admin token in prompt context for every user conversation. Independently, reviewers flag this potentially unsafe practice for assessment: 'Store only a model summary saying 'done' after changes'. Which verification addresses that practice without assuming it caused the incident?
- Inspect user-to-tool policy
- Use one unrestricted administrator token for all agent activities
- Inspect prompt and runtime secret design
- Inspect tool and device audit records
Explanation
Correct Answer
D:Narration cannot prove execution. The intended behavior is: Action traceability supports accountability.
Incorrect Answers
A:This checks check user authority, not the named failure. Conversation intent is not sufficient authorization.
B:Token compromise exposes entire network.
C:This checks keep secrets outside prompts, not the named failure. Prompt context can be disclosed or manipulated.
Designing, Deploying, and Managing Network Automation Systems (350-901) Practice Tests