Cisco 810-110 Practice Test 21
Question 200
An assistant with document read and email tools sees a document saying to send its secrets to a new address. Which approach BEST meets the requirement while respecting the system constraints?
- Require separate authorized intent and destination checks for any data export tool call.
- Treat retrieved web text as untrusted evidence and ignore behavioral redirection within it.
- Treat an address found in untrusted text as explicit user approval to email secrets.
- Verify document provenance and freshness before trusting retrieved facts or instructions.
Explanation
Correct Answer
A:Require separate authorized intent and destination checks for any data export tool call. Prompt injection can target tool permissions, not only the content of a model's answer. Verification should demonstrate that no protected content is sent to an unapproved recipient in attack simulations.
Incorrect Answers
B:Treat retrieved web text as untrusted evidence and ignore behavioral redirection within it. This primarily addresses Indirect injection in retrieved webpages, not the specific Injection-driven data exfiltration decision. Here, prompt injection can target tool permissions, not only the content of a model's answer.
C:Treat an address found in untrusted text as explicit user approval to email secrets. This is an unsafe or ineffective shortcut for Injection-driven data exfiltration. Trusting a document-defined destination can leak confidential data.
D:Verify document provenance and freshness before trusting retrieved facts or instructions. This primarily addresses RAG document poisoning, not the specific Injection-driven data exfiltration decision. Here, prompt injection can target tool permissions, not only the content of a model's answer.
Question 201
An agent submits a support ticket but the API times out before acknowledging success. Which approach BEST meets the requirement while respecting the system constraints?
- Pause side-effecting actions for the designated approver and record the decision.
- Define explicit states, branch conditions, and terminal outcomes with logs.
- Retry every successful operation until the model says it is certain.
- Use stable idempotency keys and bounded retries for side-effecting calls.
Explanation
Correct Answer
D:Use stable idempotency keys and bounded retries for side-effecting calls. At-least-once retry behavior requires duplicate suppression for nonrepeatable writes. Verification should demonstrate that replaying a request creates at most one ticket under the same business key.
Incorrect Answers
A:Pause side-effecting actions for the designated approver and record the decision. This primarily addresses Human approval gates for actions, not the specific Retry idempotency for automation decision. Here, at-least-once retry behavior requires duplicate suppression for nonrepeatable writes.
B:Define explicit states, branch conditions, and terminal outcomes with logs. This primarily addresses Multi-step workflow branching, not the specific Retry idempotency for automation decision. Here, at-least-once retry behavior requires duplicate suppression for nonrepeatable writes.
C:Retry every successful operation until the model says it is certain. This is an unsafe or ineffective shortcut for Retry idempotency for automation. Retrying without idempotency can create duplicate records and charges.
Question 202
An art workflow asks for several visual variations of one conceptual layout. Select TWO statements that correctly combine the Diffusion-based image creation method with evidence or its governing principle.
- Select an instruction-tuned language model that predicts and generates text.
- Tell a text embedding model to output its vector as a finished illustration.
- Choose a suitable text-to-image diffusion model with rights and safety checks.
- The image output reflects the scene description without copying a source image.
- Use a generative text model for drafting and a classifier when the output is a fixed category.
Explanation
Correct Answers
C:Choose a suitable text-to-image diffusion model with rights and safety checks. This is the intended action for Diffusion-based image creation.
D:The image output reflects the scene description without copying a source image. This is directly observable validation of the Diffusion-based image creation result.
Incorrect Answers
A:Select an instruction-tuned language model that predicts and generates text. This serves the different Autoregressive language generation versus image diffusion objective, not the selected Diffusion-based image creation plan. Diffusion generation iteratively denoises a sampled representation conditioned on guidance.
B:Tell a text embedding model to output its vector as a finished illustration. This is not part of a safe or effective Diffusion-based image creation design. A text-only completion model cannot natively return a high-quality generated bitmap.
E:Use a generative text model for drafting and a classifier when the output is a fixed category. This serves the different Conditional generation versus classification objective, not the selected Diffusion-based image creation plan. Diffusion generation iteratively denoises a sampled representation conditioned on guidance.
Question 203
A customer-facing assistant must answer quickly for users in one geographic region. Which approach BEST meets the requirement while respecting the system constraints?
- Estimate compute utilization, capital expense, operations, token charges, and data transfer.
- Establish an owner-run patch, version rollout, security scan, and rollback process.
- Benchmark acceptable model endpoints close to users while enforcing residency rules.
- Ignore network round trips and evaluate latency only by model parameter count.
Explanation
Correct Answer
C:Benchmark acceptable model endpoints close to users while enforcing residency rules. Hosting region, network travel, and model processing all contribute to user-visible delay. Verification should demonstrate that end-to-end p95 response latency decreases without violating location requirements.
Incorrect Answers
A:Estimate compute utilization, capital expense, operations, token charges, and data transfer. This primarily addresses Model hosting total cost of ownership, not the specific Regional hosting and latency decision. Here, hosting region, network travel, and model processing all contribute to user-visible delay.
B:Establish an owner-run patch, version rollout, security scan, and rollback process. This primarily addresses Self-hosted model updates and patching, not the specific Regional hosting and latency decision. Here, hosting region, network travel, and model processing all contribute to user-visible delay.
D:Ignore network round trips and evaluate latency only by model parameter count. This is an unsafe or ineffective shortcut for Regional hosting and latency. Selecting an endpoint solely for low token price may cause unacceptable response times.
Question 204
AI generates a responsive contact form and an interactive error banner. After the recommended UI prototype accessibility checks change is implemented, which observation MOST directly verifies that it works?
- Keyboard-only and assistive-technology users can complete the example flow.
- Existing contract tests and access checks pass for the refactored implementation.
- Requests and responses match the contract including error cases.
- Injection payloads are handled as values and unauthorized tables remain inaccessible.
Explanation
Correct Answer
A:Keyboard-only and assistive-technology users can complete the example flow. This directly tests whether the UI prototype accessibility checks goal was met. Rapid UI prototyping must include accessibility validation.
Incorrect Answers
B:Existing contract tests and access checks pass for the refactored implementation. This is evidence about Regression testing AI-proposed refactors, not direct proof of the required UI prototype accessibility checks behavior. The relevant check is that keyboard-only and assistive-technology users can complete the example flow.
C:Requests and responses match the contract including error cases. This is evidence about API scaffolding with contract-first tests, not direct proof of the required UI prototype accessibility checks behavior. The relevant check is that keyboard-only and assistive-technology users can complete the example flow.
D:Injection payloads are handled as values and unauthorized tables remain inaccessible. This is evidence about Secure generated SQL queries, not direct proof of the required UI prototype accessibility checks behavior. The relevant check is that keyboard-only and assistive-technology users can complete the example flow.
Question 205
An analyst needs consistent missing-data treatment before fitting and evaluation. Which practical correction or implementation decision is MOST directly appropriate?
- Replace all missing numbers with zero without checking what zero means.
- Fit preprocessing on training partitions and document missingness.
- Normalize with explicit parse rules and exception reporting.
- Add thresholded quality gates with source-level investigation.
Explanation
Correct Answer
B:Fit preprocessing on training partitions and document missingness. The intended behavior can be checked because test data is transformed without using hidden future or holdout statistics.
Incorrect Answers
A:Replace all missing numbers with zero without checking what zero means. This would not correct the Missing-value strategies and leakage prevention workflow. Filling validation records using statistics learned from the full dataset contaminates evaluation.
C:Normalize with explicit parse rules and exception reporting. This could help Type-aware normalization of input data, but it does not resolve the Missing-value strategies and leakage prevention case. Imputation must reflect variable semantics and avoid training-test leakage.
D:Add thresholded quality gates with source-level investigation. This could help Automated data-quality drift checks, but it does not resolve the Missing-value strategies and leakage prevention case. Imputation must reflect variable semantics and avoid training-test leakage.
Question 206
A workflow team wants agents to complete well-defined tasks rather than run forever. Testing exposes this failure: Unbounded self-reflection loops waste resources and delay delivery. Which remediation should be applied FIRST?
- Classify answer-only generation separately from agent workflows that invoke authorized actions.
- Use a bounded agent plan with tool calls, intermediate validation, and a final summary.
- Tell an agent never stop until perfect with unlimited tool calls.
- Introduce explicit stop conditions and bounded execution.
Explanation
Correct Answer
D:Introduce explicit stop conditions and bounded execution. Confirmation requires seeing that runs terminate with success, controlled failure, or approved handoff inside limits.
Incorrect Answers
A:Classify answer-only generation separately from agent workflows that invoke authorized actions. It addresses Passive text generation versus agent action, not the diagnosed Bounded autonomy versus unsupervised loops failure. Unbounded self-reflection loops waste resources and delay delivery.
B:Use a bounded agent plan with tool calls, intermediate validation, and a final summary. It addresses Goal-directed multi-step task execution, not the diagnosed Bounded autonomy versus unsupervised loops failure. Unbounded self-reflection loops waste resources and delay delivery.
C:Tell an agent never stop until perfect with unlimited tool calls. This retains the Bounded autonomy versus unsupervised loops problem. Autonomy should operate within measurable scope and constraints.
Question 207
A user asks an assistant to research three vendor options and prepare a grounded comparison. Which approach BEST meets the requirement while respecting the system constraints?
- Use a bounded agent plan with tool calls, intermediate validation, and a final summary.
- Define completion criteria, maximum steps, budget ceilings, and escalation rules.
- Call a single text-completion response verified multistep research without any tools.
- Classify answer-only generation separately from agent workflows that invoke authorized actions.
Explanation
Correct Answer
A:Use a bounded agent plan with tool calls, intermediate validation, and a final summary. Agentic flows can sequence decisions and actions toward a task goal. Verification should demonstrate that the trace shows source gathering, checked transformations, and the requested result.
Incorrect Answers
B:Define completion criteria, maximum steps, budget ceilings, and escalation rules. This primarily addresses Bounded autonomy versus unsupervised loops, not the specific Goal-directed multi-step task execution decision. Here, agentic flows can sequence decisions and actions toward a task goal.
C:Call a single text-completion response verified multistep research without any tools. This is an unsafe or ineffective shortcut for Goal-directed multi-step task execution. A single unsupported answer cannot demonstrate verified multistep execution.
D:Classify answer-only generation separately from agent workflows that invoke authorized actions. This primarily addresses Passive text generation versus agent action, not the specific Goal-directed multi-step task execution decision. Here, agentic flows can sequence decisions and actions toward a task goal.
Question 208
A multi-agent handoff requires only an order status and not full account identifiers. An engineer must explain the recommendation for Redaction during agent data transformation. Which underlying principle is the MOST accurate?
- MCP separates host coordination, client communication, and server-exposed functionality.
- Transformation stages are security boundaries where unnecessary data must be removed.
- Agents must treat external tool payloads as uncertain structured data.
- Schema mapping translates between service contracts and preserves meaning.
Explanation
Correct Answer
B:Transformation stages are security boundaries where unnecessary data must be removed. Therefore the implementation decision is to map an allowlisted minimal subset of fields and redact protected attributes.
Incorrect Answers
A:Forwarding the entire source object can expose private fields to more systems. This describes the separate MCP host-client-server roles mechanism; the present case instead requires Redaction during agent data transformation.
C:Forwarding the entire source object can expose private fields to more systems. Agents must treat external tool payloads as uncertain structured data.
D:Forwarding the entire source object can expose private fields to more systems. This describes the separate Mapping agent tool output to a target schema mechanism; the present case instead requires Redaction during agent data transformation.
Question 209
An operations lead asks which customer segments contribute to a monthly trend. Which practical correction or implementation decision is MOST directly appropriate?
- Confirm flagged records and evaluate whether anomalies reflect data errors or real events.
- Rank regions solely by absolute complaints without considering their transaction volumes.
- Add causal caveats and seek controlled or longitudinal evidence.
- Compute group-specific statistics and verify comparability.
Explanation
Correct Answer
D:Compute group-specific statistics and verify comparability. The intended behavior can be checked because segment-level tables show group denominators and comparable measures.
Incorrect Answers
A:Confirm flagged records and evaluate whether anomalies reflect data errors or real events. This could help Exploratory data anomaly identification, but it does not resolve the Segmentation of exploratory datasets case. Aggregation choices can reveal or hide patterns through grouping and denominators.
B:Rank regions solely by absolute complaints without considering their transaction volumes. This would not correct the Segmentation of exploratory datasets workflow. Comparing raw group totals without population sizes can misstate relative rates.
C:Add causal caveats and seek controlled or longitudinal evidence. This could help Correlation versus causation in AI analysis, but it does not resolve the Segmentation of exploratory datasets case. Aggregation choices can reveal or hide patterns through grouping and denominators.
Question 210
An enterprise assistant is proposed for high-stakes security access decisions. Which practical correction or implementation decision is MOST directly appropriate?
- Restrict high-impact automation and incorporate domain-qualified oversight.
- Provide meaningful AI-use disclosure and human responsibility.
- Approve emergency advice because the generated wording sounds compassionate.
- Test fairness on appropriate consented data and mitigate discovered disparities.
Explanation
Correct Answer
A:Restrict high-impact automation and incorporate domain-qualified oversight. The intended behavior can be checked because high-risk inputs trigger appropriate controls and qualified review.
Incorrect Answers
B:Provide meaningful AI-use disclosure and human responsibility. This could help Transparency about AI assistance, but it does not resolve the AI safety review for risky use cases case. Safety controls should scale with potential real-world harm.
C:Approve emergency advice because the generated wording sounds compassionate. This would not correct the AI safety review for risky use cases workflow. Releasing critical guidance without safeguards can expose users to serious consequences.
D:Test fairness on appropriate consented data and mitigate discovered disparities. This could help Bias evaluation across demographic groups, but it does not resolve the AI safety review for risky use cases case. Safety controls should scale with potential real-world harm.