McAfee Secure

Exam Code: DP-800

Exam Name: Developing AI-Enabled Database Solutions

Certification Provider: Microsoft

Corresponding Certification: Microsoft Certified: SQL AI Developer Associate

Questions & Answers

Microsoft DP-800 Questions & Answers

Study with Up-To-Date REAL Exam Questions and Answers from the ACTUAL Test

Questions & Answers

101 Questions & Answers with Testing Engine

"Developing AI-Enabled Database Solutions Exam", also known as DP-800 exam, is a Microsoft certification exam.

Pass your tests with the always up-to-date DP-800 Exam Engine. Your DP-800 training materials keep you at the head of the pack!

Satisfaction Guaranteed

Money Back Guarantee

Testking provides hassle-free money back guarantee with our products. That is because we have 100% trust in the abilities of our professional and experience product team, and our record is a proof of that.

99.6% PASS RATE
Was: $137.49
Now: $124.99

Microsoft DP-800 Practice Test 60

Question 590

Different reporting tools duplicate a sensitive sales join and occasionally disagree on calculations. Following remediation, what safeguard BEST prevents a repeat of this specific failure?

  1. Track dependent objects through versioned migrations.
  2. Control grants on base tables and treat the view as an interface contract.
  3. Verify behavior and compatibility-level differences before broad replacement.
  4. Review function ownership and data-access boundaries.

Explanation

Correct Answer

B:

Control grants on base tables and treat the view as an interface contract. This guards against recurrence because a view centralizes a query contract without storing a second copy by default. The direct remediation is: Create a governed view projecting the approved join and columns.

Incorrect Answers

A:

Track dependent objects through versioned migrations. The current task instead requires this action: control grants on base tables and treat the view as an interface contract.

C:

Verify behavior and compatibility-level differences before broad replacement. The current task instead requires this action: control grants on base tables and treat the view as an interface contract.

D:

Review function ownership and data-access boundaries. The current task instead requires this action: control grants on base tables and treat the view as an interface contract.

Question 591

A developer exposes a write-capable MCP database tool to an agent with no approval boundary. Which technical action BEST resolves the primary problem?

  1. Enforce tenant-aware authorization, bounded query complexity and back-end SQL row security.
  2. Apply dynamic data masking with appropriate UNMASK grants and test role behavior.
  3. Scope MCP tool actions and require human confirmation for consequential changes.
  4. Use targeted audit groups and evaluate performance and retention costs.

Explanation

Correct Answer

C:

Scope MCP tool actions and require human confirmation for consequential changes. This targets the primary issue because agents can turn natural-language requests into real database mutations if tools permit. Verification: A test request cannot alter billing data without approval. Prevent recurrence: Keep a tool allowlist and capture authenticated invocations.

Incorrect Answers

A:

GraphQL relationship traversal can increase data exposure if access is too broad.

B:

DDM changes what unauthorized queries display; it is not data encryption.

D:

High-volume blanket logging can impact throughput and reveal sensitive inputs.

Question 592

A container app's managed identity can log in but cannot read the required database view. Which test result BEST demonstrates that the intended solution is working?

  1. The worker connects with an approved identity and no stored SQL password.
  2. Cross-tenant test identities cannot read each other's invoice rows.
  3. A test role change creates an alertable audit record.
  4. The app reads the intended view but cannot read private financial objects.

Explanation

Correct Answer

D:

The app reads the intended view but cannot read private financial objects. This is the decisive observable evidence after implementing create or map the appropriate principal and grant minimum SQL permissions. It tests the underlying requirement: Azure identity authentication and database authorization are distinct checks.

Incorrect Answers

A:

The worker connects with an approved identity and no stored SQL password. However, that result would not establish the required outcome: the app reads the intended view but cannot read private financial objects.

B:

Cross-tenant test identities cannot read each other's invoice rows. However, that result would not establish the required outcome: the app reads the intended view but cannot read private financial objects.

C:

A test role change creates an alertable audit record. However, that result would not establish the required outcome: the app reads the intended view but cannot read private financial objects.

Question 593

A customer-support assistant must classify multilingual complaints with a small latency budget. Which technical principle BEST explains the relevant design decision?

  1. A larger model is not automatically the best fit for cost, speed or multilingual accuracy.
  2. Text-only models cannot directly interpret image content without a supporting vision pipeline.
  3. Embeddings are computed numeric representations and are not created merely by defining a VECTOR column.
  4. External model quotas and transient errors require controlled retry behavior.

Explanation

Correct Answer

A:

A larger model is not automatically the best fit for cost, speed or multilingual accuracy. This explains why the corrective action is to evaluate multilingual compact and large models on representative language and latency benchmarks. Confirm the diagnosis when blind tests meet per-language quality and p95 response targets.

Incorrect Answers

B:

Text-only models cannot directly interpret image content without a supporting vision pipeline. The actual technical constraint is that a larger model is not automatically the best fit for cost, speed or multilingual accuracy.

C:

Embeddings are computed numeric representations and are not created merely by defining a VECTOR column. The actual technical constraint is that a larger model is not automatically the best fit for cost, speed or multilingual accuracy.

D:

External model quotas and transient errors require controlled retry behavior. The actual technical constraint is that a larger model is not automatically the best fit for cost, speed or multilingual accuracy.

Question 594

A trigger on SalesLine inserts assumes only one row appears in inserted and updates one summary incorrectly. Following remediation, what safeguard BEST prevents a repeat of this specific failure?

  1. Implement transactional outbox or supported change notifications.
  2. Use multirow tests and avoid cursor-based per-row assumptions.
  3. Verify behavior and compatibility-level differences before broad replacement.
  4. Track dependent objects through versioned migrations.

Explanation

Correct Answer

B:

Use multirow tests and avoid cursor-based per-row assumptions. This guards against recurrence because SQL triggers run per statement and inserted can contain multiple rows. The direct remediation is: Rewrite the DML trigger to handle the inserted/deleted pseudo-tables as sets.

Incorrect Answers

A:

Implement transactional outbox or supported change notifications. The current task instead requires this action: use multirow tests and avoid cursor-based per-row assumptions.

C:

Verify behavior and compatibility-level differences before broad replacement. The current task instead requires this action: use multirow tests and avoid cursor-based per-row assumptions.

D:

Track dependent objects through versioned migrations. The current task instead requires this action: use multirow tests and avoid cursor-based per-row assumptions.

Question 595

An app selects an enormous model for every short product tag and then exceeds budget. What is the MOST appropriate immediate corrective approach?

  1. Prefer a model and API mode supporting validated structured output and enforce schema checks.
  2. Configure change tracking to refresh embeddings every time a product record is edited.
  3. Benchmark smaller models and route complex cases only when quality needs justify it.
  4. Include semantically meaningful title and content fields, with identifiers retained as metadata.

Explanation

Correct Answer

C:

Benchmark smaller models and route complex cases only when quality needs justify it. This targets the primary issue because model size increases cost and latency without guaranteeing proportional quality gains. Verification: The tiered approach meets held-out accuracy with lower cost per accepted result. Prevent recurrence: Track model quality drift and escalation thresholds.

Incorrect Answers

A:

Unconstrained prose is fragile to parse and can omit required keys.

B:

Change tracking can improve vector freshness, but it does not reduce the cost of using an oversized inference model for each short product tag.

D:

IDs alone lack the natural-language signal needed for semantic retrieval.

Question 596

A billing schema is owned by finance but anyone in the repository can approve SQL changes. What observed evidence would BEST confirm successful remediation?

  1. Unapproved branch commits never reach production.
  2. Scans find no credentials and a staging deployment authenticates securely.
  3. Repository scans show no live credentials while seed records remain deterministic.
  4. A billing SQL PR requires a finance-code-owner approval before merge.

Explanation

Correct Answer

D:

A billing SQL PR requires a finance-code-owner approval before merge. This is the decisive observable evidence after implementing configure CODEOWNERS and required reviews for the finance database paths. It tests the underlying requirement: Path ownership helps route high-impact DDL for accountable review.

Incorrect Answers

A:

Unapproved branch commits never reach production. However, that result would not establish the required outcome: a billing SQL PR requires a finance-code-owner approval before merge.

B:

Scans find no credentials and a staging deployment authenticates securely. However, that result would not establish the required outcome: a billing SQL PR requires a finance-code-owner approval before merge.

C:

Repository scans show no live credentials while seed records remain deterministic. However, that result would not establish the required outcome: a billing SQL PR requires a finance-code-owner approval before merge.

Question 597

A junior engineer accepts an AI-generated DROP TABLE migration without checking dependencies. What is the MOST relevant technical distinction in this situation?

  1. AI suggestions are proposals, not change-control authorization.
  2. AI chat providers and extensions may retain shared content outside the database boundary.
  3. Copilot in Fabric availability can depend on region, capacity and tenant policies.
  4. An MCP client needs a discoverable trusted server and permissions to run tools.

Explanation

Correct Answer

A:

AI suggestions are proposals, not change-control authorization. This explains why the corrective action is to review the generated diff, run it in an isolated test database, and require approval. Confirm the diagnosis when migration rehearsal retains required tables and supports rollback.

Incorrect Answers

B:

AI chat providers and extensions may retain shared content outside the database boundary. The actual technical constraint is that AI suggestions are proposals, not change-control authorization.

C:

Copilot in Fabric availability can depend on region, capacity and tenant policies. The actual technical constraint is that AI suggestions are proposals, not change-control authorization.

D:

An MCP client needs a discoverable trusted server and permissions to run tools. The actual technical constraint is that AI suggestions are proposals, not change-control authorization.

Question 598

An Azure SQL procedure needs to send a controlled JSON prompt to an approved external model API. Which ongoing control is MOST relevant after the issue is fixed?

  1. Rotate permissions and avoid hardcoded bearer tokens.
  2. Use managed identity/secure credentials and enforce timeouts.
  3. Audit access metadata and citation validity.
  4. Test corpus freshness and unauthorized access.

Explanation

Correct Answer

B:

Use managed identity/secure credentials and enforce timeouts. This guards against recurrence because the stored procedure enables outbound HTTP but requires correct permissions, endpoint reachability and payload formatting. The direct remediation is: Build the request payload and call sp_invoke_external_rest_endpoint through a supported secure configuration.

Incorrect Answers

A:

Rotate permissions and avoid hardcoded bearer tokens. The current task instead requires this action: use managed identity/secure credentials and enforce timeouts.

C:

Audit access metadata and citation validity. The current task instead requires this action: use managed identity/secure credentials and enforce timeouts.

D:

Test corpus freshness and unauthorized access. The current task instead requires this action: use managed identity/secure credentials and enforce timeouts.

Question 599

A supplier feed sometimes sends malformed JSON in an otherwise valid text column. What is the MOST appropriate immediate corrective approach?

  1. Use OPENJSON with a typed WITH schema and transactional parent-child inserts.
  2. Replace FLOAT money arithmetic with appropriately sized DECIMAL precision and scale.
  3. Validate JSON structure on ingestion using supported native JSON type or ISJSON checks where applicable.
  4. Design a useful composite rowstore index with leading keys matching the filters and inspect the execution plan.

Explanation

Correct Answer

C:

Validate JSON structure on ingestion using supported native JSON type or ISJSON checks where applicable. This targets the primary issue because malformed payloads can break downstream JSON_VALUE and OPENJSON processing. Verification: Invalid payloads are rejected into an error queue without corrupting accepted rows. Prevent recurrence: Keep the supplier contract and dead-letter handling versioned.

Incorrect Answers

A:

Parsing nested arrays into typed rows reduces fragile manual string extraction.

B:

Binary floating-point approximations cannot represent every decimal currency value exactly.

D:

A missing selective access path forces large table or clustered-index scans.

Question 600

A publicly exposed DAB instance floods SQL with unauthenticated expensive read requests. What observed evidence would BEST confirm successful remediation?

  1. Container-based smoke tests connect securely to the intended Azure SQL database.
  2. An authorized request runs the report and rejects unauthorized procedure calls.
  3. A staging start resolves each declared source without runtime errors.
  4. Load tests enforce limits and approved users retain service.

Explanation

Correct Answer

D:

Load tests enforce limits and approved users retain service. This is the decisive observable evidence after implementing harden ingress, authentication, rate limiting and query bounds before production exposure. It tests the underlying requirement: Autoscaling the API does not eliminate downstream database exhaustion or unauthorized access.

Incorrect Answers

A:

Container-based smoke tests connect securely to the intended Azure SQL database. However, that result would not establish the required outcome: load tests enforce limits and approved users retain service.

B:

An authorized request runs the report and rejects unauthorized procedure calls. However, that result would not establish the required outcome: load tests enforce limits and approved users retain service.

C:

A staging start resolves each declared source without runtime errors. However, that result would not establish the required outcome: load tests enforce limits and approved users retain service.

Developing AI-Enabled Database Solutions (DP-800) Practice Tests