cert
cert-1
cert-2

Pass Microsoft AI-103 Exam in First Attempt Guaranteed!

Get 100% Latest Exam Questions, Accurate & Verified Answers to Pass the Actual Exam!
30 Days Free Updates, Instant Download!

cert-5
cert-6
AI-103 Exam - Verified By Experts
AI-103 Premium File

AI-103 Premium File

$89.99
$98.99
  • Premium File 187 Questions & Answers. Last Update: Nov 15, 2026

Whats Included:

  • Latest Questions
  • 100% Accurate Answers
  • Fast Exam Updates
 
$98.99
$89.99
accept 397 downloads in the last 7 days
block-screenshots
Microsoft AI-103 Practice Test 45

Question 440

A form extractor confidently outputs a claim amount that does not match the printed total. What release check should be enforced?

  1. Validate extracted amounts against source evidence and business rules before downstream writes
  2. Write extracted claim amounts immediately without checking arithmetic or source evidence
  3. Produce supported Markdown or text representations retaining structure and source lineage
  4. Configure a Content Understanding analyzer with task-specific structured output fields

Explanation

Correct Answer

A:

Extraction confidence does not replace checks against the actual source and numeric invariants. The practical requirement is to verify extracted values before financial side effects; a solution built to analyze mixed-media documents and assets would not meet that same need.

Incorrect Answers

B:

A confident extraction can still contradict printed totals. Validate monetary values against the document and business rules before a consequential write.

C:

Readable grounded representations allow reviewers and RAG systems to inspect supporting document content. The function serves to prepare structured readable document evidence, whereas the actual requirement is to verify extracted values before financial side effects.

D:

Defined extraction fields make downstream claim validation and automation dependable. That capability is appropriate for produce consistent structured outputs from claim documents, not this separate task of finding how to verify extracted values before financial side effects.

Question 441

A model improves with one-shot examples but fails on unrepresented edge cases. What prompt-testing practice is necessary?

  1. Lock a prompt after one polished demonstration without running edge-case and adversarial evaluations
  2. Specify task priority and an ordered output schema for separate subrequests
  3. Use supported schema-constrained output and deterministic post-parse validation
  4. Evaluate diverse examples and hard negatives before locking the prompt template

Explanation

Correct Answer

D:

Prompt examples can bias the model toward narrow patterns, so edge-case coverage matters. Here the relevant capability must let the team validate prompt behavior on diverse difficult cases. An alternative targeting reduce creative variation in structured summaries solves another problem.

Incorrect Answers

A:

One example is not representative of the deployment distribution. Evaluate varied inputs and negative cases before treating the prompt as stable.

B:

Clear task decomposition reduces ambiguity when one input contains multiple decisions. The function serves to organize multi-intent generation deterministically, whereas the actual requirement is to validate prompt behavior on diverse difficult cases.

C:

Required keys and types should be enforced with a machine-readable contract, not example text alone. That capability is appropriate for produce parseable fixed-field JSON, not this separate task of finding how to validate prompt behavior on diverse difficult cases.

Question 442

A consumer agent reads hostile web pages and can modify account data. Which THREE measures reduce realistic misuse? Choose THREE.

  1. Rewrite refusal messages to sound friendlier without changing tool authorization.
  2. Remove sensitive data from the final reply but allow instructions embedded in hostile pages to execute account-changing tools without authorization.
  3. Trust commands embedded in hostile web pages whenever a generated-content provenance marker accompanies them.
  4. Treat retrieved document instructions as untrusted data and enforce tool permissions
  5. Validate tool arguments and authorize every side effect on the server
  6. Use a policy-driven high-risk response gate with human escalation

Explanation

Correct Answers

D:

Indirect prompt injection attempts to turn lower-trust evidence into commands; tool authorization must not depend on retrieved prose. Untrusted page content must never override developer policy. Tool arguments and account-write permissions must be checked at the server boundary, with high-risk operations routed through an explicit human-review gate. Pleasant refusal wording without enforcement does not mitigate the write risk.

E:

A prompt-only instruction is not an access-control mechanism; programmatic checks constrain sensitive writes.

F:

A high-risk approval gate prevents sensitive account actions from proceeding solely on the model’s interpretation of a hostile document.

Incorrect Answers

A:

Warmer wording alone cannot prevent indirect prompt injection or unauthorized writes through the available tools.

B:

Output redaction cannot stop a malicious page from causing unauthorized writes. Tool authorization, untrusted-input boundaries and high-risk approval gates address that failure.

C:

Page content remains untrusted even when labeled. Enforce tool authorization, schema validation and human gates for high-risk actions.

Question 443

An image moderation model overblocks harmless educational diagrams. How should the team improve?

  1. Calibrate visual safety thresholds on reviewed allowed and prohibited examples
  2. Apply brand-usage policy checks and require approved assets before publication
  3. Apply clear provenance labeling or watermarking where appropriate
  4. Ignore the false positives and only block instructions found by OCR in the educational diagrams.

Explanation

Correct Answer

A:

Moderation requires measuring both overblocking and missed harmful content. This resolves the specific need to reduce visual safety false positives, while a measure that helps enforce approved trademark usage in generated media addresses a different requirement.

Incorrect Answers

B:

Visual generation does not override intellectual-property or brand governance requirements. That is relevant to enforce approved trademark usage in generated media, but the present scenario requires an ability to reduce visual safety false positives.

C:

Synthetic-media disclosure reduces confusion over whether visuals show real products or events. This helps when a team needs to disclose generated-media provenance; it does not establish how to reduce visual safety false positives.

D:

Embedded-text filtering does not calibrate moderation errors. Evaluate the classifier on representative permitted educational material and adjust policy thresholds or human-review paths based on measured false-positive rates.

Question 444

A global assistant may receive text, speech or images and must keep a common conversation history. What data representation helps?

  1. Store only a flat string transcript that discards which turns contained images, audio or timestamps.
  2. Ask for explicit confirmation of the recognized command before side effects
  3. Normalize modality-specific evidence into tagged multimodal conversation turns
  4. Preserve task context while updating the input and output language preference

Explanation

Correct Answer

C:

Tagged input blocks preserve source and modality while maintaining ordered task context. The decisive difference is whether the design can integrate different input modalities in one conversation; the competing function serves to detect completion of a spoken user turn.

Incorrect Answers

A:

Losing modality and temporal markers prevents accurate multimodal replay. Conversation state should retain typed turns and evidence references for each input modality.

B:

Voice recognition errors can be costly, so important tool actions require verified user intent. Here the needed outcome is to integrate different input modalities in one conversation; an option designed to confirm uncertain spoken commands before actions would not directly provide it.

D:

Language changes should not reset authorized task state or leak previous user's context. That is relevant to handle mid-session language changes safely, but the present scenario requires an ability to integrate different input modalities in one conversation.

Question 445

A response cites a document but a reviewer cannot determine which revision it came from. What index metadata is essential?

  1. Cite only the current document title, without a revision identifier or source hash, when reproducing an earlier answer.
  2. Remove or invalidate corresponding vectors and cached excerpts during deletion handling
  3. Apply source ACL and tenant authorization at retrieval time
  4. Store source version identifiers with retrieved chunks and citations

Explanation

Correct Answer

D:

Document version lineage enables auditing which exact evidence informed the response. Here the relevant capability must let the team identify referenced document revisions. An alternative targeting attach traceable evidence to generated answers solves another problem.

Incorrect Answers

A:

Document titles may remain identical across edits. Source version or digest metadata must accompany retrieved chunks and cited spans to identify the actual evidence edition.

B:

Derived retrieval artifacts must follow source retention and deletion requirements. The function serves to remove expired source-derived vectors and snippets, whereas the actual requirement is to identify referenced document revisions.

C:

A shared vector index must not bypass permissions present on the authoritative documents. That capability is appropriate for prevent restricted documents crossing retrieval boundaries, not this separate task of finding how to identify referenced document revisions.

Question 446

A customer service bot answers invoices in several currencies and changes the decimal separator inconsistently. What validation helps?

  1. Use locale-aware formatting and numeric business-rule checks after model generation.
  2. Apply one decimal separator to every country without checking currency locale.
  3. Remove currency codes so formatted amounts look shorter.
  4. Trust that any parseable number is valid even when it differs from the invoice.

Explanation

Correct Answer

A:

Localization must preserve numeric meaning, not merely fluent wording, particularly for money.

Incorrect Answers

B:

Uniform formatting can misread or misstate amounts when a locale uses a different decimal or grouping rule.

C:

Dropping currency codes makes amounts ambiguous even if their decimal separators look consistent. Validate locale-specific formatting while preserving an explicit currency identifier and exact numeric value.

D:

Syntactic parsing cannot guarantee source-value fidelity or valid monetary rules.

Question 447

Quality analysts need to know whether a model's summary is supported by a provided case file. What evaluator should be emphasized?

  1. Grade summaries solely by fluency and formatting, without comparing claims to the actual case file.
  2. Run privacy-preserving subgroup performance and fairness checks on appropriate test data
  3. Provide traceable source references and decision-factor provenance for review
  4. Use groundedness or faithfulness evaluation against the source document

Explanation

Correct Answer

D:

A fluent summary can still invent claims; groundedness tests whether assertions are supported by the reference. The decisive difference is whether the design can score source-supported generated statements; the competing function serves to evaluate whether answers actually fulfill the task.

Incorrect Answers

A:

A polished but unsupported summary can score well on style while hallucinating facts. Groundedness evaluation must compare stated claims with the source evidence.

B:

Aggregate subgroup metrics can expose disparate performance without publishing individual records. Here the needed outcome is to score source-supported generated statements; an option designed to check fairness across relevant cohorts would not directly provide it.

C:

Auditable evidence and factor summaries explain why the system recommended an outcome without exposing private hidden reasoning. Its primary purpose is to make AI-supported decisions explainable with evidence; the stated constraint instead demands a way to score source-supported generated statements.

Question 448

A public outreach campaign uses AI-generated video footage that could be mistaken for documentary evidence. What provenance measure helps viewers?

  1. Publish the generated video without disclosure because it depicts a plausible event.
  2. Delete generation metadata before distribution to make the footage look authentic.
  3. Apply appropriate synthetic-media disclosure and verifiable provenance metadata where supported.
  4. Use only video compression quality as the approval criterion.

Explanation

Correct Answer

C:

Content labels and provenance help prevent misleading presentation of generated media as real event footage.

Incorrect Answers

A:

Lifelike synthetic footage may be mistaken for documentary evidence without provenance labels.

B:

Removing provenance weakens transparency and can increase the likelihood of misleading viewers.

D:

Encoding quality does not address whether synthetic footage is presented misleadingly.

Question 449

A caller switches from German voice to English text mid-session. What state behavior should the agent implement?

  1. Preserve task context while updating the input and output language preference
  2. Reset conversation state whenever the customer switches from voice to text
  3. Use endpoint detection and turn-taking signals before final response synthesis
  4. Ask for explicit confirmation of the recognized command before side effects

Explanation

Correct Answer

A:

Language changes should not reset authorized task state or leak previous user's context. In this scenario the team must handle mid-session language changes safely. A feature used to integrate different input modalities in one conversation does not substitute for it.

Incorrect Answers

B:

A modality change does not cancel the original task. Preserve authorized context while updating the language and input channel.

C:

Voice interaction needs explicit speech-turn boundaries to avoid interrupting or mishearing requests. That is relevant to detect completion of a spoken user turn, but the present scenario requires an ability to handle mid-session language changes safely.

D:

Voice recognition errors can be costly, so important tool actions require verified user intent. This helps when a team needs to confirm uncertain spoken commands before actions; it does not establish how to handle mid-session language changes safely.

Question 450

An agent retrieves a restricted HR policy because the index is shared across teams. What must be enforced?

  1. Use a shared search index as permission to display HR passages to every department
  2. Apply source ACL and tenant authorization at retrieval time
  3. Remove or invalidate corresponding vectors and cached excerpts during deletion handling
  4. Store source version identifiers with retrieved chunks and citations

Explanation

Correct Answer

B:

A shared vector index must not bypass permissions present on the authoritative documents. This directly supports the need to prevent restricted documents crossing retrieval boundaries, unlike a mechanism primarily intended to attach traceable evidence to generated answers.

Incorrect Answers

A:

Co-location in an index is not an authorization grant. Retrieval must enforce the source's ACL and tenant scope.

C:

Derived retrieval artifacts must follow source retention and deletion requirements. Here the needed outcome is to prevent restricted documents crossing retrieval boundaries; an option designed to remove expired source-derived vectors and snippets would not directly provide it.

D:

Document version lineage enables auditing which exact evidence informed the response. Its primary purpose is to identify referenced document revisions; the stated constraint instead demands a way to prevent restricted documents crossing retrieval boundaries.

Developing AI Apps and Agents on Azure (AI-103) Practice Tests