cert
cert-1
cert-2

Pass Microsoft AI-300 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-300 Exam - Verified By Experts
AI-300 Premium File

AI-300 Premium File

$69.99
$76.99
  • Premium File 213 Questions & Answers. Last Update: Nov 12, 2026

Whats Included:

  • Latest Questions
  • 100% Accurate Answers
  • Fast Exam Updates
 
$76.99
$69.99
accept 45 downloads in the last 7 days
block-screenshots
Microsoft AI-300 Practice Test 78

Question 770

During a Foundry generative AI quality and observability incident, the following condition occurs: An evaluation compares chatbot answers against an automatically rewritten query rather than the original user intent, rewarding responses that miss the real troubleshooting question. The fix must preserve evaluation validity and avoid unnecessary user-data collection. What should the engineer examine first to confirm the specific AI-300 workload issue?

  1. Examine supported language list, safety evaluation examples, category coverage, severity thresholds and live red-team evidence.
  2. Examine evaluation records, chat history serialization, tool traces, single-turn versus multi-turn settings and representative failure conversations.
  3. Review user input field mapping, generated search query, response text, evaluator prompt and evaluation row samples.
  4. Review usage logs for chat and embedding deployments, prompt/output token categories, query frequency, indexing cycles and resource billing tags.

Explanation

Correct Answer

C:

Review user input field mapping, generated search query, response text, evaluator prompt and evaluation row samples. This is the most relevant evidence for relevance evaluator uses wrong intended user query. Query rewriting aids retrieval but should not silently redefine what the user's response needs to accomplish. The matching controlled remediation is: Bind relevance evaluation to the authentic user request and assess query rewriting as a separate retrieval-stage metric. This matters here because an evaluation compares chatbot answers against an automatically rewritten query rather than the original user….

Incorrect Answers

A:

Examine supported language list, safety evaluation examples, category coverage, severity thresholds and live red-team evidence. This would investigate harmful content probe set excludes multilingual input rather than the documented relevance evaluator uses wrong intended user query issue. The relevant investigation is: Review user input field mapping, generated search query, response text, evaluator prompt and evaluation row samples.

B:

Examine evaluation records, chat history serialization, tool traces, single-turn versus multi-turn settings and representative failure conversations. This would investigate multi-turn agent evaluation uses isolated last message only rather than the documented relevance evaluator uses wrong intended user query issue. The relevant investigation is: Review user input field mapping, generated search query, response text, evaluator prompt and evaluation row samples.

D:

Review usage logs for chat and embedding deployments, prompt/output token categories, query frequency, indexing cycles and resource billing tags. This would investigate token cost report omits embedding requests rather than the documented relevance evaluator uses wrong intended user query issue. The relevant investigation is: Review user input field mapping, generated search query, response text, evaluator prompt and evaluation row samples.

Question 771

During a regulated Azure ML workspace infrastructure review, the following condition occurs: A model retraining pipeline silently picks up new records every week because its input references an Azure ML data asset's latest version rather than the approved immutable snapshot. The team must preserve approved access, model provenance and private network constraints. What approved correction resolves the failure without modifying unrelated healthy services?

  1. Give the approved compute or workspace identity the necessary ACR pull access and use a validated immutable image reference.
  2. Pin the pipeline input to an approved versioned data asset and advance versions through controlled evaluation gates.
  3. Declare proper resource references or explicit dependencies and staged network prerequisites in the reviewed Bicep deployment.
  4. Rebuild or correct the versioned MLTable asset transformation so the expected columns and types are supplied to downstream training.

Explanation

Correct Answer

B:

Pin the pipeline input to an approved versioned data asset and advance versions through controlled evaluation gates. This targets versionless dataset reference breaks reproducibility. A data asset reference without explicit versioning can change the effective training corpus between runs and invalidate reproducible model comparisons. Verify the failure through: Inspect the data asset version resolution, pipeline input expression, snapshot timestamps, storage content hashes and previous run provenance. Validate success with: Replaying a prior run uses identical input version and produces traceable comparison metrics before a new snapshot is promoted. This matters here because a model retraining pipeline silently picks up new records every week because its input references an Azure ML data….

Incorrect Answers

A:

Give the approved compute or workspace identity the necessary ACR pull access and use a validated immutable image reference. This could address docker environment image cannot pull from private registry, but it does not solve the specified versionless dataset reference breaks reproducibility failure. The matching remediation is: Pin the pipeline input to an approved versioned data asset and advance versions through controlled evaluation gates.

C:

Declare proper resource references or explicit dependencies and staged network prerequisites in the reviewed Bicep deployment. This could address bicep workspace declaration misses dependent private resources, but it does not solve the specified versionless dataset reference breaks reproducibility failure. The matching remediation is: Pin the pipeline input to an approved versioned data asset and advance versions through controlled evaluation gates.

D:

Rebuild or correct the versioned MLTable asset transformation so the expected columns and types are supplied to downstream training. This could address mltable data asset schema mismatches training code, but it does not solve the specified versionless dataset reference breaks reproducibility failure. The matching remediation is: Pin the pipeline input to an approved versioned data asset and advance versions through controlled evaluation gates. This matters here because a model retraining pipeline silently picks up new records every week because its input references an Azure ML data….

Question 772

While optimizing a production RAG or fine-tuned model service, the following condition occurs: A customer support model fine-tuned on historical chats becomes inconsistent because examples mix strict JSON tool calls with casual prose for identical task types. The resulting model must meet held-out quality and security criteria. What approved correction resolves the failure without modifying unrelated healthy services?

  1. Retain governed immutable fine-tune artifacts and record deployment lineage with tested restore/rollback procedures.
  2. Clean and stratify the supervised fine-tuning data to a coherent supported task format before retraining and evaluating.
  3. Expand representative domain pairs and negatives, apply appropriate embedding model adaptation and retain a separate general-language retrieval evaluation.
  4. Rebalance curated synthetic and real cases to preserve rare operational scenarios and evaluate on an untouched real-world edge-case set.

Explanation

Correct Answer

B:

Clean and stratify the supervised fine-tuning data to a coherent supported task format before retraining and evaluating. This targets supervised fine-tuning dataset includes conflicting assistant behaviors. Supervised fine-tuning learns patterns in labeled examples; contradictory targets can degrade reliability even when many examples are provided. Verify the failure through: Inspect fine-tuning records, role/message formatting, target response schemas, label consistency and curated holdout behavior. Validate success with: The customized model returns the approved structured behavior on held-out scenarios without a spike in schema violations. This matters here because while optimizing a production RAG or fine-tuned model service, the following condition occurs: A customer support….

Incorrect Answers

A:

Retain governed immutable fine-tune artifacts and record deployment lineage with tested restore/rollback procedures. This could address fine-tune production endpoint references deleted training artifact, but it does not solve the specified supervised fine-tuning dataset includes conflicting assistant behaviors failure. The matching remediation is: Clean and stratify the supervised fine-tuning data to a coherent supported task format before retraining and evaluating. This matters here because while optimizing a production RAG or fine-tuned model service, the following condition occurs: A customer support….

C:

Expand representative domain pairs and negatives, apply appropriate embedding model adaptation and retain a separate general-language retrieval evaluation. This could address specialized embedding adaptation overfits small terminology set, but it does not solve the specified supervised fine-tuning dataset includes conflicting assistant behaviors failure. The matching remediation is: Clean and stratify the supervised fine-tuning data to a coherent supported task format before retraining and evaluating.

D:

Rebalance curated synthetic and real cases to preserve rare operational scenarios and evaluate on an untouched real-world edge-case set. This could address synthetic data overwhelms minority real-world edge cases, but it does not solve the specified supervised fine-tuning dataset includes conflicting assistant behaviors failure. The matching remediation is: Clean and stratify the supervised fine-tuning data to a coherent supported task format before retraining and evaluating.

Question 773

During a regulated Azure ML workspace infrastructure review, the following condition occurs: An organization disables public access on Azure ML backing storage, but its private compute subnet cannot resolve the storage account private endpoint and training jobs time out. The team must preserve approved access, model provenance and private network constraints. What should the engineer examine first to confirm the specific AI-300 workload issue?

  1. Inspect MLTable path definitions, transformation instructions, sampled file headers, declared feature schema and job parsing errors.
  2. Inspect model metadata, job source snapshot, repository history, CI release manifest and available recorded commit identifiers.
  3. Inspect storage private endpoint approvals, privatelink DNS A records, compute subnet DNS, routes and outbound access test logs.
  4. Inspect Git branch protection, pull-request review requirements, test workflow triggers, dataset schema checks and recent commit history.

Explanation

Correct Answer

C:

Inspect storage private endpoint approvals, privatelink DNS A records, compute subnet DNS, routes and outbound access test logs. This is the most relevant evidence for aml private workspace storage access fails from compute. Private endpoints require correct name resolution and route reachability; disabling public access without these dependencies can strand model training workloads. The matching controlled remediation is: Link the correct private DNS zone and approved storage private endpoint to the managed network path before enforcing storage isolation.

Incorrect Answers

A:

Inspect MLTable path definitions, transformation instructions, sampled file headers, declared feature schema and job parsing errors. This would investigate mltable data asset schema mismatches training code rather than the documented aml private workspace storage access fails from compute issue. The relevant investigation is: Inspect storage private endpoint approvals, privatelink DNS A records, compute subnet DNS, routes and outbound access test logs.

B:

Inspect model metadata, job source snapshot, repository history, CI release manifest and available recorded commit identifiers. This would investigate production training script commit cannot be traced rather than the documented aml private workspace storage access fails from compute issue. The relevant investigation is: Inspect storage private endpoint approvals, privatelink DNS A records, compute subnet DNS, routes and outbound access test logs.

D:

Inspect Git branch protection, pull-request review requirements, test workflow triggers, dataset schema checks and recent commit history. This would investigate notebook merge bypasses protected evaluation tests rather than the documented aml private workspace storage access fails from compute issue. The relevant investigation is: Inspect storage private endpoint approvals, privatelink DNS A records, compute subnet DNS, routes and outbound access test logs.

Question 774

During a Foundry generative AI quality and observability incident, the following condition occurs: A Foundry quality monitor aggregates responses from old and new agent deployments, masking a groundedness regression in the new version. The fix must preserve evaluation validity and avoid unnecessary user-data collection. What measured outcome validates the solution while preserving the stated constraints?

  1. An adversarial document cannot cause unapproved transactions while authorized research questions still succeed.
  2. A canary regression is visible independently of stable baseline traffic before expanding rollout.
  3. The service meets per-minute token and tail-latency targets under representative large-context requests.
  4. The responsible stage is identified and P95 user response time improves without gratuitous changes to healthy components.

Explanation

Correct Answer

B:

A canary regression is visible independently of stable baseline traffic before expanding rollout. Aggregate monitoring can hide release-specific failures; quality assessment needs traceable version attribution. a direct corrective action is to segment continuous evaluation by model/agent deployment version and configure a focused regression alert for the candidate. This matters here because a Foundry quality monitor aggregates responses from old and new agent deployments, masking a groundedness regression….

Incorrect Answers

A:

An adversarial document cannot cause unapproved transactions while authorized research questions still succeed. That addresses a different configuration issue rather than the main requirement. Verify instead: A canary regression is visible independently of stable baseline traffic before expanding rollout.

C:

The service meets per-minute token and tail-latency targets under representative large-context requests. That addresses a different configuration issue rather than the main requirement. Verify instead: A canary regression is visible independently of stable baseline traffic before expanding rollout.

D:

The responsible stage is identified and P95 user response time improves without gratuitous changes to healthy components. That addresses a different configuration issue rather than the main requirement. Verify instead: A canary regression is visible independently of stable baseline traffic before expanding rollout.

Question 775

During a Foundry generative AI quality and observability incident, the following condition occurs: A retrieval assistant's monthly bill grows, but its cost dashboard counts only chat completion tokens and omits repeated query and document embedding operations. The fix must preserve evaluation validity and avoid unnecessary user-data collection. Which TWO complementary choices identify the issue and correct it? Select TWO.

  1. Mask sensitive tokens and personal data, apply least-privilege log access and retain only necessary structured debugging evidence.
  2. Correct the test dataset mapping for query, context, response and reference fields and validate a sampled evaluation before the full run.
  3. Examine end-to-end request traces, retrieval timing, model time to first token, generation duration, request concurrency and percentiles.
  4. Review usage logs for chat and embedding deployments, prompt/output token categories, query frequency, indexing cycles and resource billing tags.
  5. Instrument cost attribution across generation and embedding services, cache reusable embeddings and alert on unplanned spend.

Explanation

Correct Answers

D:

Review usage logs for chat and embedding deployments, prompt/output token categories, query frequency, indexing cycles and resource billing tags. This is the most relevant evidence for token cost report omits embedding requests. GenAI cost analysis must include all model and supporting resources; tracking only final generation underestimates retrieval and indexing spend. The matching controlled remediation is: Instrument cost attribution across generation and embedding services, cache reusable embeddings and alert on unplanned spend. This matters here because a retrieval assistant's monthly bill grows, but its cost dashboard counts only chat completion tokens and omits….

E:

Instrument cost attribution across generation and embedding services, cache reusable embeddings and alert on unplanned spend. This targets token cost report omits embedding requests. GenAI cost analysis must include all model and supporting resources; tracking only final generation underestimates retrieval and indexing spend. Verify the failure through: Review usage logs for chat and embedding deployments, prompt/output token categories, query frequency, indexing cycles and resource billing tags. Validate success with: The report reconciles billed service usage with per-feature consumption and unnecessary repeat embedding calls decline. This matters here because a retrieval assistant's monthly bill grows, but its cost dashboard counts only chat completion tokens and omits….

Incorrect Answers

A:

Mask sensitive tokens and personal data, apply least-privilege log access and retain only necessary structured debugging evidence. This is associated with debug traces leak secrets and retrieved personal records, not the essential evidence-and-fix pair for token cost report omits embedding requests. First examine: Review usage logs for chat and embedding deployments, prompt/output token categories, query frequency, indexing cycles and resource billing tags. Then correct: Instrument cost attribution across generation and embedding services, cache reusable embeddings and alert on unplanned spend.

B:

Correct the test dataset mapping for query, context, response and reference fields and validate a sampled evaluation before the full run. This is associated with genai evaluation dataset lacks expected answer column mapping, not the essential evidence-and-fix pair for token cost report omits embedding requests. First examine: Review usage logs for chat and embedding deployments, prompt/output token categories, query frequency, indexing cycles and resource billing tags. Then correct: Instrument cost attribution across generation and embedding services, cache reusable embeddings and alert on unplanned spend.

C:

Examine end-to-end request traces, retrieval timing, model time to first token, generation duration, request concurrency and percentiles. This is associated with generation latency spike blamed on retrieval without breakdown, not the essential evidence-and-fix pair for token cost report omits embedding requests. First examine: Review usage logs for chat and embedding deployments, prompt/output token categories, query frequency, indexing cycles and resource billing tags. Then correct: Instrument cost attribution across generation and embedding services, cache reusable embeddings and alert on unplanned spend.

Question 776

During a retrieval or model-customization performance review, the following condition occurs: A synthetic data workflow for complaint classification injects real support tickets verbatim as prompts and emits generated examples still containing customer names and account details. The change must keep citations, privacy and previously approved behavior intact. Which diagnostic evidence most directly distinguishes the described AI operations failure?

  1. Examine synthetic-generation input provenance, PII detection scans, privacy transformations, output similarity to real records and review approvals.
  2. Examine fine-tuning run outputs, model registry versions, promotion approvals, held-out evaluations and rollout routing settings.
  3. Compare query and index embedding model versions, vector dimensions, index schema, retrieval score distributions and migration logs.
  4. Inspect fine-tuning records, role/message formatting, target response schemas, label consistency and curated holdout behavior.

Explanation

Correct Answer

A:

Examine synthetic-generation input provenance, PII detection scans, privacy transformations, output similarity to real records and review approvals. This is the most relevant evidence for synthetic training examples copy private customer identifiers. Generated data is not automatically privacy-safe; a model can reproduce sensitive source details when the generation prompt includes them. The matching controlled remediation is: Generate sanitized synthetic patterns from approved nonidentifying templates and screen outputs for memorized or copied private content. This matters here because a synthetic data workflow for complaint classification injects real support tickets verbatim as prompts and emits….

Incorrect Answers

B:

Examine fine-tuning run outputs, model registry versions, promotion approvals, held-out evaluations and rollout routing settings. This would investigate fine-tuned candidate bypasses promotion review after training rather than the documented synthetic training examples copy private customer identifiers issue. The relevant investigation is: Examine synthetic-generation input provenance, PII detection scans, privacy transformations, output similarity to real records and review approvals.

C:

Compare query and index embedding model versions, vector dimensions, index schema, retrieval score distributions and migration logs. This would investigate embedding model replaced without reindexing stored vectors rather than the documented synthetic training examples copy private customer identifiers issue. The relevant investigation is: Examine synthetic-generation input provenance, PII detection scans, privacy transformations, output similarity to real records and review approvals.

D:

Inspect fine-tuning records, role/message formatting, target response schemas, label consistency and curated holdout behavior. This would investigate supervised fine-tuning dataset includes conflicting assistant behaviors rather than the documented synthetic training examples copy private customer identifiers issue. The relevant investigation is: Examine synthetic-generation input provenance, PII detection scans, privacy transformations, output similarity to real records and review approvals.

Question 777

During a production Azure Machine Learning model lifecycle incident, the following condition occurs: An Azure ML training pipeline intermittently starts training before its feature-engineering output is ready because the training component references a static path instead of the upstream component's declared output. The operator must safeguard training reproducibility and serving continuity. Select TWO actions: one decisive diagnostic and its corresponding targeted remediation.

  1. Check pipeline YAML, component inputs and outputs, resource dependency graph and job execution chronology.
  2. Bind training input directly to the feature component's typed output to establish the correct pipeline dependency.
  3. Inspect sweep max_concurrent_trials, compute cluster maximum nodes, GPU family quota and job scheduling telemetry.
  4. Register a supported MLflow model with validated flavor, explicit input signature, pinned dependencies and a successful clean-environment load test.
  5. Enable supported model artifact logging or explicitly log a serializable MLflow model with its dependencies and signature.

Explanation

Correct Answers

A:

Check pipeline YAML, component inputs and outputs, resource dependency graph and job execution chronology. This is the most relevant evidence for pipeline graph omits feature-build dependency. Component dependency edges in Azure ML are determined by declared data bindings, not by visual position in a notebook or YAML file. The matching controlled remediation is: Bind training input directly to the feature component's typed output to establish the correct pipeline dependency. This matters here because during a production Azure Machine Learning model lifecycle incident, the following condition occurs: An Azure ML….

B:

Bind training input directly to the feature component's typed output to establish the correct pipeline dependency. This targets pipeline graph omits feature-build dependency. Component dependency edges in Azure ML are determined by declared data bindings, not by visual position in a notebook or YAML file. Verify the failure through: Check pipeline YAML, component inputs and outputs, resource dependency graph and job execution chronology. Validate success with: The feature job completes before training consumes its versioned output in repeated orchestrated executions. This matters here because during a production Azure Machine Learning model lifecycle incident, the following condition occurs: An Azure ML….

Incorrect Answers

C:

Inspect sweep max_concurrent_trials, compute cluster maximum nodes, GPU family quota and job scheduling telemetry. This is associated with sweep concurrency overwhelms scarce gpu quota, not the essential evidence-and-fix pair for pipeline graph omits feature-build dependency. First examine: Check pipeline YAML, component inputs and outputs, resource dependency graph and job execution chronology. Then correct: Bind training input directly to the feature component's typed output to establish the correct pipeline dependency.

D:

Register a supported MLflow model with validated flavor, explicit input signature, pinned dependencies and a successful clean-environment load test. This is associated with mlflow registry artifact cannot load on serving target, not the essential evidence-and-fix pair for pipeline graph omits feature-build dependency. First examine: Check pipeline YAML, component inputs and outputs, resource dependency graph and job execution chronology. Then correct: Bind training input directly to the feature component's typed output to establish the correct pipeline dependency.

E:

Enable supported model artifact logging or explicitly log a serializable MLflow model with its dependencies and signature. This is associated with autologging excludes required model artifact, not the essential evidence-and-fix pair for pipeline graph omits feature-build dependency. First examine: Check pipeline YAML, component inputs and outputs, resource dependency graph and job execution chronology. Then correct: Bind training input directly to the feature component's typed output to establish the correct pipeline dependency.

Question 778

During a Microsoft Foundry generative AI deployment review, the following condition occurs: A prototype assistant runs only a few tests per day but reserves a large provisioned-throughput deployment, creating a high recurring bill even when no requests arrive. The deployment must retain governed identities, endpoint contracts and auditable release history. Which production-safe remediation should the engineer apply to this AI workflow?

  1. Permit only the required documented tool and service destinations via supported private or managed-network egress controls.
  2. Compare available candidate models on representative multilingual datasets and select one that meets the target-language quality and security gates.
  3. Right-size the supported provisioned capacity or apply controlled request pacing and overflow strategy within budget and region limits.
  4. Move the low-volume prototype to an appropriate supported consumption-based deployment or reduce provisioned capacity after review.

Explanation

Correct Answer

D:

Move the low-volume prototype to an appropriate supported consumption-based deployment or reduce provisioned capacity after review. This targets provisioned deployment pays for idle unused capacity. Provisioned throughput may be billed for allocated capacity independent of actual use; deployment type should match utilization patterns. Verify the failure through: Examine billed PTU hours, token utilization, workload predictability, available pay-as-you-go endpoints and expected launch load. Validate success with: The prototype maintains acceptable response time with lower recurring idle capacity cost and production capacity remains unaffected. This matters here because a prototype assistant runs only a few tests per day but reserves a large provisioned-throughput deployment, creating….

Incorrect Answers

A:

Permit only the required documented tool and service destinations via supported private or managed-network egress controls. This could address agent tool calls fail behind overbroad outbound block, but it does not solve the specified provisioned deployment pays for idle unused capacity failure. The matching remediation is: Move the low-volume prototype to an appropriate supported consumption-based deployment or reduce provisioned capacity after review.

B:

Compare available candidate models on representative multilingual datasets and select one that meets the target-language quality and security gates. This could address multilingual model selection fails target-language quality, but it does not solve the specified provisioned deployment pays for idle unused capacity failure. The matching remediation is: Move the low-volume prototype to an appropriate supported consumption-based deployment or reduce provisioned capacity after review.

C:

Right-size the supported provisioned capacity or apply controlled request pacing and overflow strategy within budget and region limits. This could address provisioned throughput capacity insufficient at peak traffic, but it does not solve the specified provisioned deployment pays for idle unused capacity failure. The matching remediation is: Move the low-volume prototype to an appropriate supported consumption-based deployment or reduce provisioned capacity after review.

Question 779

As an MLOps engineer reviews a model training or inference pipeline, the following condition occurs: A real-time predictor appears healthy by mean request latency while high-value users face multi-second waits at the 99th percentile after a traffic spike. The resolution must preserve quality gates, documented artifacts and rollback options. Which acceptance test shows the AI operations defect is corrected?

  1. P95/P99 and error-rate service objectives recover under representative peak-load traffic without unrelated scale changes.
  2. The feature job completes before training consumes its versioned output in repeated orchestrated executions.
  3. A poor candidate is retained for review without reaching production, and qualifying candidates progress through controlled canary rollout.
  4. An injected null-rate increase triggers a targeted alert while valid fluctuations remain within control limits.

Explanation

Correct Answer

A:

P95/P99 and error-rate service objectives recover under representative peak-load traffic without unrelated scale changes. Average endpoint latency can hide slow outliers; tail distributions and cohort behavior are needed for user-facing service assurance. a direct corrective action is to configure tail-latency and error-rate dashboards with capacity remedies appropriate to the measured bottleneck. This matters here because as an MLOps engineer reviews a model training or inference pipeline, the following condition occurs: A real-time….

Incorrect Answers

B:

The feature job completes before training consumes its versioned output in repeated orchestrated executions. That addresses a different configuration issue rather than the main requirement. Verify instead: P95/P99 and error-rate service objectives recover under representative peak-load traffic without unrelated scale changes.

C:

A poor candidate is retained for review without reaching production, and qualifying candidates progress through controlled canary rollout. That addresses a different configuration issue rather than the main requirement. Verify instead: P95/P99 and error-rate service objectives recover under representative peak-load traffic without unrelated scale changes.

D:

An injected null-rate increase triggers a targeted alert while valid fluctuations remain within control limits. That addresses a different configuration issue rather than the main requirement. Verify instead: P95/P99 and error-rate service objectives recover under representative peak-load traffic without unrelated scale changes. This matters here because as an MLOps engineer reviews a model training or inference pipeline, the following condition occurs: A real-time….

Question 780

While preparing a secure Foundry model or agent release, the following condition occurs: An API service rolls back prompt text after a deployment incident but leaves the incompatible new foundation-model version in place, so tool-call formatting errors persist. The remediation must meet enterprise access and reliability requirements without a broad bypass. Which diagnostic evidence most directly distinguishes the described AI operations failure?

  1. Inspect deployed model modalities, model card capabilities, payload content types, supported API version and failed inference responses.
  2. Audit Foundry network ACLs, private endpoint connections, VNet peerings, firewall rules and effective caller identities.
  3. Inspect input/output token volume, peak concurrent requests, provisioned throughput allocation, throttling codes and deployment latency percentiles.
  4. Compare prompt commit hashes, foundation model version, tool JSON schema, serving configuration and prior approved release bundle.

Explanation

Correct Answer

D:

Compare prompt commit hashes, foundation model version, tool JSON schema, serving configuration and prior approved release bundle. This is the most relevant evidence for prompt contract rollback ignores associated model version. A generative application behavior depends on versioned combinations of model, prompts and tools, not any single text template. The matching controlled remediation is: Restore the complete compatible model, prompt and tool schema release manifest and retest its evaluated workflow. This matters here because while preparing a secure Foundry model or agent release, the following condition occurs: An API service rolls back….

Incorrect Answers

A:

Inspect deployed model modalities, model card capabilities, payload content types, supported API version and failed inference responses. This would investigate vision input sent to text-only foundation model rather than the documented prompt contract rollback ignores associated model version issue. The relevant investigation is: Compare prompt commit hashes, foundation model version, tool JSON schema, serving configuration and prior approved release bundle. This matters here because while preparing a secure Foundry model or agent release, the following condition occurs: An API service rolls back….

B:

Audit Foundry network ACLs, private endpoint connections, VNet peerings, firewall rules and effective caller identities. This would investigate network allowlist exposes production model endpoint to test subnet rather than the documented prompt contract rollback ignores associated model version issue. The relevant investigation is: Compare prompt commit hashes, foundation model version, tool JSON schema, serving configuration and prior approved release bundle.

C:

Inspect input/output token volume, peak concurrent requests, provisioned throughput allocation, throttling codes and deployment latency percentiles. This would investigate provisioned throughput capacity insufficient at peak traffic rather than the documented prompt contract rollback ignores associated model version issue. The relevant investigation is: Compare prompt commit hashes, foundation model version, tool JSON schema, serving configuration and prior approved release bundle.

Operationalizing Machine Learning and Generative AI Solutions (AI-300) Practice Tests