Cisco 810-110 Practice Test 54
Question 530
A team needs to choose which requests justify slower deliberate inference. Select TWO statements that correctly combine the Reasoning versus speed trade-off method with evidence or its governing principle.
- Route routine tasks to a lightweight model and difficult reasoning to a validated capable model.
- Quality and p95 latency meet separate targets for each request class.
- Choose maximum-reasoning settings for every short greeting without measurement.
- Review model-card license, permitted uses, redistribution obligations, and provenance.
- Inspect model cards for supported input/output modalities and test the required format.
Explanation
Correct Answers
A:Route routine tasks to a lightweight model and difficult reasoning to a validated capable model. This is the intended action for Reasoning versus speed trade-off.
B:Quality and p95 latency meet separate targets for each request class. This is directly observable validation of the Reasoning versus speed trade-off result.
Incorrect Answers
C:Choose maximum-reasoning settings for every short greeting without measurement. This is not part of a safe or effective Reasoning versus speed trade-off design. Sending every trivial request through the slowest expensive reasoning model raises latency and spend.
D:Review model-card license, permitted uses, redistribution obligations, and provenance. This serves the different Model license and usage restrictions objective, not the selected Reasoning versus speed trade-off plan. Different model capability and cost profiles support differentiated routing.
E:Inspect model cards for supported input/output modalities and test the required format. This serves the different Modality-aware model repository selection objective, not the selected Reasoning versus speed trade-off plan. Different model capability and cost profiles support differentiated routing.
Question 531
An engineer wants a quick query builder that must resist SQL injection. Testing exposes this failure: String-concatenated generated SQL can allow attacker-controlled commands. Which remediation should be applied FIRST?
- Build raw SQL by concatenating user input because the model wrote the code.
- Use parameters and security tests instead of interpolation.
- Generate API scaffolding from the approved contract and test endpoints against it.
- Provide acceptance tests first and iterate generated code until they pass.
Explanation
Correct Answer
B:Use parameters and security tests instead of interpolation. Confirmation requires seeing that injection payloads are handled as values and unauthorized tables remain inaccessible.
Incorrect Answers
A:Build raw SQL by concatenating user input because the model wrote the code. This retains the Secure generated SQL queries problem. Secure query parameter binding separates data from executable SQL text.
C:Generate API scaffolding from the approved contract and test endpoints against it. It addresses API scaffolding with contract-first tests, not the diagnosed Secure generated SQL queries failure. String-concatenated generated SQL can allow attacker-controlled commands.
D:Provide acceptance tests first and iterate generated code until they pass. It addresses Test-driven AI code generation, not the diagnosed Secure generated SQL queries failure. String-concatenated generated SQL can allow attacker-controlled commands.
Question 532
A generated Python service raises a TypeError on a data path. After the recommended Debugging from stack traces with AI change is implemented, which observation MOST directly verifies that it works?
- Findings are reviewed and policy-breaking defects are blocked.
- Operators can isolate the failing stage without leaking secret payloads.
- The original failing test passes without regressing unaffected behavior.
- The implementation preserves invariants across a broad generated test set.
Explanation
Correct Answer
C:The original failing test passes without regressing unaffected behavior. This directly tests whether the Debugging from stack traces with AI goal was met. Error evidence narrows debugging hypotheses but requires execution to confirm.
Incorrect Answers
A:Findings are reviewed and policy-breaking defects are blocked. This is evidence about Static-analysis and lint-assisted review, not direct proof of the required Debugging from stack traces with AI behavior. The relevant check is that the original failing test passes without regressing unaffected behavior.
B:Operators can isolate the failing stage without leaking secret payloads. This is evidence about Observability for AI workflows, not direct proof of the required Debugging from stack traces with AI behavior. The relevant check is that the original failing test passes without regressing unaffected behavior.
D:The implementation preserves invariants across a broad generated test set. This is evidence about Edge-case and property-based tests, not direct proof of the required Debugging from stack traces with AI behavior. The relevant check is that the original failing test passes without regressing unaffected behavior.
Question 533
A regulated European business is evaluating inference endpoints in several jurisdictions. After the recommended Geographic data residency approval change is implemented, which observation MOST directly verifies that it works?
- High-risk systems receive enhanced tests, human review, and monitoring.
- Network and storage checks verify cryptographic protections and limited decryption rights.
- Data-flow inventories and contracts show processing stays within approved jurisdictions.
- Restricted documents are absent for users without permission in negative tests.
Explanation
Correct Answer
C:Data-flow inventories and contracts show processing stays within approved jurisdictions. This directly tests whether the Geographic data residency approval goal was met. Residency can include telemetry, storage, and backups, not only the primary API endpoint.
Incorrect Answers
A:High-risk systems receive enhanced tests, human review, and monitoring. This is evidence about AI use-case risk classification, not direct proof of the required Geographic data residency approval behavior. The relevant check is that data-flow inventories and contracts show processing stays within approved jurisdictions.
B:Network and storage checks verify cryptographic protections and limited decryption rights. This is evidence about Encryption and key management of AI data, not direct proof of the required Geographic data residency approval behavior. The relevant check is that data-flow inventories and contracts show processing stays within approved jurisdictions.
D:Restricted documents are absent for users without permission in negative tests. This is evidence about Enterprise AI connector DLP enforcement, not direct proof of the required Geographic data residency approval behavior. The relevant check is that data-flow inventories and contracts show processing stays within approved jurisdictions.
Question 534
A team wants suggestions for readability without treating AI as the sole gate. An engineer must explain the recommendation for AI support for code review. Which underlying principle is the MOST accurate?
- AI reviews are advisory and can miss context-specific defects.
- Static tools identify classes of errors that fluent model output may hide.
- AI can structure requirements but cannot authorize undocumented business assumptions.
- Prototyping reduces uncertainty but is not equivalent to production readiness.
Explanation
Correct Answer
A:AI reviews are advisory and can miss context-specific defects. Therefore the implementation decision is to use AI code-review suggestions alongside human inspection and executable checks.
Incorrect Answers
B:Automatically merging on a favorable model summary can miss security and logic flaws. This describes the separate Static-analysis and lint-assisted review mechanism; the present case instead requires AI support for code review.
C:Automatically merging on a favorable model summary can miss security and logic flaws. AI can structure requirements but cannot authorize undocumented business assumptions.
D:Automatically merging on a favorable model summary can miss security and logic flaws. Prototyping reduces uncertainty but is not equivalent to production readiness.
Question 535
A reviewer observes that postal codes drive unexplained differences in recommendations. Which practical correction or implementation decision is MOST directly appropriate?
- Declare a model fair solely because its training file has no protected-attribute column.
- Review proxies, measure impact, and adjust model inputs and policy.
- Include diverse users and assistive workflows in evaluation.
- Provide meaningful AI-use disclosure and human responsibility.
Explanation
Correct Answer
B:Review proxies, measure impact, and adjust model inputs and policy. The intended behavior can be checked because tests identify whether indirect features cause unjustified performance or outcome gaps.
Incorrect Answers
A:Declare a model fair solely because its training file has no protected-attribute column. This would not correct the Proxy variable fairness risk workflow. Hidden proxy variables can reproduce bias despite removing sensitive labels.
C:Include diverse users and assistive workflows in evaluation. This could help Inclusive accessibility testing, but it does not resolve the Proxy variable fairness risk case. Excluding an explicit protected column does not eliminate proxy discrimination.
D:Provide meaningful AI-use disclosure and human responsibility. This could help Transparency about AI assistance, but it does not resolve the Proxy variable fairness risk case. Excluding an explicit protected column does not eliminate proxy discrimination.
Question 536
A search-enabled assistant reads a webpage containing instructions to ignore the original research request. After the recommended Indirect injection in retrieved webpages change is implemented, which observation MOST directly verifies that it works?
- Answers cite authenticated approved policies instead of injected forged versions.
- Injected directives inside excerpts are treated as content rather than commands.
- Tests show lower-priority messages cannot replace protected configuration.
- The assistant answers the user's question without executing page-authored commands.
Explanation
Correct Answer
D:The assistant answers the user's question without executing page-authored commands. This directly tests whether the Indirect injection in retrieved webpages goal was met. Indirect prompt injection hides control attempts inside lower-trust material.
Incorrect Answers
A:Answers cite authenticated approved policies instead of injected forged versions. This is evidence about RAG document poisoning, not direct proof of the required Indirect injection in retrieved webpages behavior. The relevant check is that the assistant answers the user's question without executing page-authored commands.
B:Injected directives inside excerpts are treated as content rather than commands. This is evidence about Delimiting external context in prompts, not direct proof of the required Indirect injection in retrieved webpages behavior. The relevant check is that the assistant answers the user's question without executing page-authored commands.
C:Tests show lower-priority messages cannot replace protected configuration. This is evidence about Direct user instruction override attempt, not direct proof of the required Indirect injection in retrieved webpages behavior. The relevant check is that the assistant answers the user's question without executing page-authored commands.
Question 537
A planning assistant must retain agreed constraints across a long interaction. An engineer must explain the recommendation for Conversation memory summarization. Which underlying principle is the MOST accurate?
- Termination controls bound output but do not replace semantic completeness checks.
- Persistent task state should preserve decisions and provenance while limiting token growth.
- Long context capacity does not guarantee equal attention to every position.
- Similarity search compares embedding geometry; it does not guarantee factual truth.
Explanation
Correct Answer
B:Persistent task state should preserve decisions and provenance while limiting token growth. Therefore the implementation decision is to keep a verified compact state summary plus recent turns and retrievable references.
Incorrect Answers
A:Naively discarding old turns can lose binding customer preferences. Termination controls bound output but do not replace semantic completeness checks.
C:Naively discarding old turns can lose binding customer preferences. Long context capacity does not guarantee equal attention to every position.
D:Naively discarding old turns can lose binding customer preferences. Similarity search compares embedding geometry; it does not guarantee factual truth.
Question 538
A regulated team has a hard policy forbidding patient records from leaving an approved internal network. If the relevant Private on-premises model deployment safeguard is omitted, which outcome is the MOST directly associated failure?
- Deploying untested generated code can introduce runtime and security defects.
- Selecting an endpoint solely for low token price may cause unacceptable response times.
- Comparing only subscription sticker prices obscures hardware and labor expenses.
- Sending prohibited records to a public model API breaches the declared data boundary.
Explanation
Correct Answer
D:Sending prohibited records to a public model API breaches the declared data boundary. Local hosting can reduce external data transfer but adds infrastructure and patching duties. The corrective approach is to deploy and maintain an approved private model with controlled telemetry.
Incorrect Answers
A:Deploying untested generated code can introduce runtime and security defects. This failure pattern belongs chiefly to Code-generation model with execution tests; it is not the failure most directly tested in the Private on-premises model deployment situation. Sending prohibited records to a public model API breaches the declared data boundary.
B:Selecting an endpoint solely for low token price may cause unacceptable response times. This failure pattern belongs chiefly to Regional hosting and latency; it is not the failure most directly tested in the Private on-premises model deployment situation. Sending prohibited records to a public model API breaches the declared data boundary.
C:Comparing only subscription sticker prices obscures hardware and labor expenses. This failure pattern belongs chiefly to Model hosting total cost of ownership; it is not the failure most directly tested in the Private on-premises model deployment situation. Sending prohibited records to a public model API breaches the declared data boundary.
Question 539
A workflow team wants agents to complete well-defined tasks rather than run forever. Which practical correction or implementation decision is MOST directly appropriate?
- Specify whether each workflow is advisory or action-capable and enforce permissions.
- Tell an agent never stop until perfect with unlimited tool calls.
- Introduce explicit stop conditions and bounded execution.
- Orchestrate planned steps with checks and reviewable outputs.
Explanation
Correct Answer
C:Introduce explicit stop conditions and bounded execution. The intended behavior can be checked because runs terminate with success, controlled failure, or approved handoff inside limits.
Incorrect Answers
A:Specify whether each workflow is advisory or action-capable and enforce permissions. This could help Passive text generation versus agent action, but it does not resolve the Bounded autonomy versus unsupervised loops case. Autonomy should operate within measurable scope and constraints.
B:Tell an agent never stop until perfect with unlimited tool calls. This would not correct the Bounded autonomy versus unsupervised loops workflow. Unbounded self-reflection loops waste resources and delay delivery.
D:Orchestrate planned steps with checks and reviewable outputs. This could help Goal-directed multi-step task execution, but it does not resolve the Bounded autonomy versus unsupervised loops case. Autonomy should operate within measurable scope and constraints.
Question 540
An analyst asks for a statement of fact backed by source passages. Which practical correction or implementation decision is MOST directly appropriate?
- Apply evidence sufficiency checks and abstain when unsupported.
- Match citations to retrieved evidence and flag unverifiable statements.
- Validate source references and decline unsupported details.
- Add fake-looking source identifiers to every model claim regardless of evidence.
Explanation
Correct Answer
C:Validate source references and decline unsupported details. The intended behavior can be checked because each supported assertion maps to an actual source passage and unknowns are disclosed.
Incorrect Answers
A:Apply evidence sufficiency checks and abstain when unsupported. This could help Appropriate uncertainty and abstention, but it does not resolve the Grounded answers with citations case. Attribution reduces unsupported claims but citations must be checked for correctness.
B:Match citations to retrieved evidence and flag unverifiable statements. This could help Separation of prompt guidance from evidence verification, but it does not resolve the Grounded answers with citations case. Attribution reduces unsupported claims but citations must be checked for correctness.
D:Add fake-looking source identifiers to every model claim regardless of evidence. This would not correct the Grounded answers with citations workflow. Fabricated citations can make hallucinated answers appear authoritative.