Cisco 810-110 Practice Test 28
Question 270
A document-processing agent recursively calls an expensive model while retrying failures. Which approach BEST meets the requirement while respecting the system constraints?
- Increase retry concurrency without limit whenever an API reports throttling.
- Set concurrency, retry, token, and spending budgets with circuit breakers.
- Define explicit states, branch conditions, and terminal outcomes with logs.
- Design compensating actions or resumable checkpoints around side effects.
Explanation
Correct Answer
B:Set concurrency, retry, token, and spending budgets with circuit breakers. AI workflow capacity needs explicit budget and backpressure controls. Verification should demonstrate that run metrics show bounded model usage without exhausting service quotas.
Incorrect Answers
A:Increase retry concurrency without limit whenever an API reports throttling. This is an unsafe or ineffective shortcut for Rate limits and cost budgets. Unbounded loops can produce runaway spend and cascading throttling.
C:Define explicit states, branch conditions, and terminal outcomes with logs. This primarily addresses Multi-step workflow branching, not the specific Rate limits and cost budgets decision. Here, aI workflow capacity needs explicit budget and backpressure controls.
D:Design compensating actions or resumable checkpoints around side effects. This primarily addresses Workflow compensation after partial failure, not the specific Rate limits and cost budgets decision. Here, aI workflow capacity needs explicit budget and backpressure controls.
Question 271
A company finds a downloadable checkpoint in a public model hub for commercial embedding. Which approach BEST meets the requirement while respecting the system constraints?
- Treat a public download button as automatic waiver of all obligations.
- Pin available model version identifiers and log parameters, prompt, and evaluation dataset.
- Inspect model cards for supported input/output modalities and test the required format.
- Review model-card license, permitted uses, redistribution obligations, and provenance.
Explanation
Correct Answer
D:Review model-card license, permitted uses, redistribution obligations, and provenance. Availability in a repository does not mean unrestricted commercial rights. Verification should demonstrate that legal review records that the intended deployment meets the artifact terms.
Incorrect Answers
A:Treat a public download button as automatic waiver of all obligations. This is an unsafe or ineffective shortcut for Model license and usage restrictions. Assuming every open-weight model is public domain can violate license terms.
B:Pin available model version identifiers and log parameters, prompt, and evaluation dataset. This primarily addresses Reproducible model version pinning, not the specific Model license and usage restrictions decision. Here, availability in a repository does not mean unrestricted commercial rights.
C:Inspect model cards for supported input/output modalities and test the required format. This primarily addresses Modality-aware model repository selection, not the specific Model license and usage restrictions decision. Here, availability in a repository does not mean unrestricted commercial rights.
Question 272
An internal AI assistant can search documents and invoke production APIs. After the recommended Role-based access for AI service tools change is implemented, which observation MOST directly verifies that it works?
- Data-flow inventories and contracts show processing stays within approved jurisdictions.
- Network and storage checks verify cryptographic protections and limited decryption rights.
- High-risk inputs trigger appropriate controls and qualified review.
- Attempted unauthorized tool calls are denied and logged by backend enforcement.
Explanation
Correct Answer
D:Attempted unauthorized tool calls are denied and logged by backend enforcement. This directly tests whether the Role-based access for AI service tools goal was met. Least-privilege authorization must be enforced outside natural-language instructions.
Incorrect Answers
A:Data-flow inventories and contracts show processing stays within approved jurisdictions. This is evidence about Geographic data residency approval, not direct proof of the required Role-based access for AI service tools behavior. The relevant check is that attempted unauthorized tool calls are denied and logged by backend enforcement.
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 Role-based access for AI service tools behavior. The relevant check is that attempted unauthorized tool calls are denied and logged by backend enforcement.
C:High-risk inputs trigger appropriate controls and qualified review. This is evidence about AI safety review for risky use cases, not direct proof of the required Role-based access for AI service tools behavior. The relevant check is that attempted unauthorized tool calls are denied and logged by backend enforcement.
Question 273
A multilingual support app has measurable network delay to a distant model endpoint. Which practical correction or implementation decision is MOST directly appropriate?
- Ignore network round trips and evaluate latency only by model parameter count.
- Benchmark regional endpoints and choose a compliant low-latency path.
- Build a realistic workload cost model before selecting a hosting option.
- Select a hosted service and set throughput and budget controls.
Explanation
Correct Answer
B:Benchmark regional endpoints and choose a compliant low-latency path. The intended behavior can be checked because end-to-end p95 response latency decreases without violating location requirements.
Incorrect Answers
A:Ignore network round trips and evaluate latency only by model parameter count. This would not correct the Regional hosting and latency workflow. Selecting an endpoint solely for low token price may cause unacceptable response times.
C:Build a realistic workload cost model before selecting a hosting option. This could help Model hosting total cost of ownership, but it does not resolve the Regional hosting and latency case. Hosting region, network travel, and model processing all contribute to user-visible delay.
D:Select a hosted service and set throughput and budget controls. This could help Managed cloud inference elasticity, but it does not resolve the Regional hosting and latency case. Hosting region, network travel, and model processing all contribute to user-visible delay.
Question 274
A researcher needs a summary spanning many reports that exceed single-call context. After the recommended Long-document chunk and merge change is implemented, which observation MOST directly verifies that it works?
- The merged summary cites source chunks and covers all required sections.
- Responses stop at expected boundaries and report explicit truncation status.
- A complete response validates against the required schema without truncation.
- Legal review records that the intended deployment meets the artifact terms.
Explanation
Correct Answer
A:The merged summary cites source chunks and covers all required sections. This directly tests whether the Long-document chunk and merge goal was met. Map-reduce summarization trades extra steps for bounded context and source traceability.
Incorrect Answers
B:Responses stop at expected boundaries and report explicit truncation status. This is evidence about Stop conditions and maximum output, not direct proof of the required Long-document chunk and merge behavior. The relevant check is that the merged summary cites source chunks and covers all required sections.
C:A complete response validates against the required schema without truncation. This is evidence about Response output token reservation, not direct proof of the required Long-document chunk and merge behavior. The relevant check is that the merged summary cites source chunks and covers all required sections.
D:Legal review records that the intended deployment meets the artifact terms. This is evidence about Model license and usage restrictions, not direct proof of the required Long-document chunk and merge behavior. The relevant check is that the merged summary cites source chunks and covers all required sections.
Question 275
A developer uses AI to propose unit tests for a price calculation routine. After the recommended Unit test planning in SDLC change is implemented, which observation MOST directly verifies that it works?
- Generated code meets the test cases including boundaries and invalid inputs.
- Deterministic tests pass and fail when a seeded defect is introduced.
- Stakeholders confirm the stories cover their real needs and constraints.
- Failed health metrics stop promotion or trigger recovery to a known-good release.
Explanation
Correct Answer
B:Deterministic tests pass and fail when a seeded defect is introduced. This directly tests whether the Unit test planning in SDLC goal was met. AI may help design tests while execution establishes observed behavior.
Incorrect Answers
A:Generated code meets the test cases including boundaries and invalid inputs. This is evidence about Test-driven AI code generation, not direct proof of the required Unit test planning in SDLC behavior. The relevant check is that deterministic tests pass and fail when a seeded defect is introduced.
C:Stakeholders confirm the stories cover their real needs and constraints. This is evidence about AI-assisted requirements elicitation, not direct proof of the required Unit test planning in SDLC behavior. The relevant check is that deterministic tests pass and fail when a seeded defect is introduced.
D:Failed health metrics stop promotion or trigger recovery to a known-good release. This is evidence about Deployment gates and rollback planning, not direct proof of the required Unit test planning in SDLC behavior. The relevant check is that deterministic tests pass and fail when a seeded defect is introduced.
Question 276
A team collected data for customer support but wants to reuse private messages for unrelated profiling. Which approach BEST meets the requirement while respecting the system constraints?
- Assume any accessible dataset can be reused for any future AI experiment.
- Supply a least-privilege sanitized sandbox dataset with controlled export.
- Apply minimum cohort size rules and suppress identifying small cells.
- Verify permitted purpose, notices, and lawful basis before secondary AI use.
Explanation
Correct Answer
D:Verify permitted purpose, notices, and lawful basis before secondary AI use. Ethical AI analysis respects the context and permissions under which data was collected. Verification should demonstrate that approval records show the new processing purpose is allowed.
Incorrect Answers
A:Assume any accessible dataset can be reused for any future AI experiment. This is an unsafe or ineffective shortcut for Purpose limitation and consent in analysis. Reusing data outside approved purposes can violate trust and contractual constraints.
B:Supply a least-privilege sanitized sandbox dataset with controlled export. This primarily addresses Secure analytics sandboxing, not the specific Purpose limitation and consent in analysis decision. Here, ethical AI analysis respects the context and permissions under which data was collected.
C:Apply minimum cohort size rules and suppress identifying small cells. This primarily addresses Privacy-preserving aggregated reporting, not the specific Purpose limitation and consent in analysis decision. Here, ethical AI analysis respects the context and permissions under which data was collected.
Question 277
A search index includes public and restricted chunks with owner metadata. An engineer must explain the recommendation for Retrieval ACL and metadata filtering. Which underlying principle is the MOST accurate?
- RAG brings external evidence into a model context without requiring weight retraining.
- RAG indexes are derived copies that require lifecycle synchronization.
- Vector proximity does not override document or tenant access controls.
- Persistent task state should preserve decisions and provenance while limiting token growth.
Explanation
Correct Answer
C:Vector proximity does not override document or tenant access controls. Therefore the implementation decision is to apply per-user authorization and metadata filters before exposing retrieved context.
Incorrect Answers
A:Unguarded nearest-neighbor retrieval can leak private records into generated answers. RAG brings external evidence into a model context without requiring weight retraining.
B:Unguarded nearest-neighbor retrieval can leak private records into generated answers. This describes the separate Retrieval index freshness mechanism; the present case instead requires Retrieval ACL and metadata filtering.
D:Unguarded nearest-neighbor retrieval can leak private records into generated answers. Persistent task state should preserve decisions and provenance while limiting token growth.
Question 278
A public dataset is modified to associate a trusted phrase with a malicious response. After the recommended Training data poisoning change is implemented, which observation MOST directly verifies that it works?
- Reviewers can trace who approved what and which data affected the action.
- Audited reviews cite primary telemetry rather than treating model prose as proof.
- Only approved real dependencies enter builds after independent checks.
- Poisoned samples are detected or excluded before training and evaluation.
Explanation
Correct Answer
D:Poisoned samples are detected or excluded before training and evaluation. This directly tests whether the Training data poisoning goal was met. Corrupted training examples can steer learned behavior or embed backdoors.
Incorrect Answers
A:Reviewers can trace who approved what and which data affected the action. This is evidence about Traceable AI decision audit logs, not direct proof of the required Training data poisoning behavior. The relevant check is that poisoned samples are detected or excluded before training and evaluation.
B:Audited reviews cite primary telemetry rather than treating model prose as proof. This is evidence about Automation bias and overreliance, not direct proof of the required Training data poisoning behavior. The relevant check is that poisoned samples are detected or excluded before training and evaluation.
C:Only approved real dependencies enter builds after independent checks. This is evidence about AI-generated dependency security risks, not direct proof of the required Training data poisoning behavior. The relevant check is that poisoned samples are detected or excluded before training and evaluation.
Question 279
A model produces a training-record fragment when prompted with a rare identifier. If the relevant Sensitive data memorization exposure safeguard is omitted, which outcome is the MOST directly associated failure?
- Treating training data as impossible to recover creates privacy risk.
- A source document may make a harmless summarization task perform exfiltration.
- Publishing unverified generated content can amplify misinformation and reputational harm.
- Unreported model incidents can persist without correction or customer notice.
Explanation
Correct Answer
A:Treating training data as impossible to recover creates privacy risk. Models can memorize and emit fragments of sensitive training data. The corrective approach is to limit sensitive training content and test leakage with governed access.
Incorrect Answers
B:A source document may make a harmless summarization task perform exfiltration. This failure pattern belongs chiefly to Tool-enabled prompt injection security; it is not the failure most directly tested in the Sensitive data memorization exposure situation. Treating training data as impossible to recover creates privacy risk.
C:Publishing unverified generated content can amplify misinformation and reputational harm. This failure pattern belongs chiefly to Synthetic misinformation and impersonation; it is not the failure most directly tested in the Sensitive data memorization exposure situation. Treating training data as impossible to recover creates privacy risk.
D:Unreported model incidents can persist without correction or customer notice. This failure pattern belongs chiefly to Policy enforcement and incident response; it is not the failure most directly tested in the Sensitive data memorization exposure situation. Treating training data as impossible to recover creates privacy risk.
Question 280
A documentation prompt must identify its audience, task, and permitted evidence. Select TWO statements that correctly combine the Clear instruction role and context method with evidence or its governing principle.
- Define the expected output fields, types, allowed values, and validate parsed results.
- Supply only the word helpful and assume the assistant infers every requirement.
- State the role, precise task, audience, source context, and output boundaries.
- Test cases return appropriately scoped answers in the required voice and format.
- Keep safety and application rules in higher-priority instructions and reject conflicting lower-priority requests.
Explanation
Correct Answers
C:State the role, precise task, audience, source context, and output boundaries. This is the intended action for Clear instruction role and context.
D:Test cases return appropriately scoped answers in the required voice and format. This is directly observable validation of the Clear instruction role and context result.
Incorrect Answers
A:Define the expected output fields, types, allowed values, and validate parsed results. This serves the different Explicit output schema constraints objective, not the selected Clear instruction role and context plan. Explicit goals and constraints reduce ambiguity without granting the model real-world authority.
B:Supply only the word helpful and assume the assistant infers every requirement. This is not part of a safe or effective Clear instruction role and context design. Vague requests can produce irrelevant output or omit required constraints.
E:Keep safety and application rules in higher-priority instructions and reject conflicting lower-priority requests. This serves the different System versus user instruction priorities objective, not the selected Clear instruction role and context plan. Explicit goals and constraints reduce ambiguity without granting the model real-world authority.