Question 1000A model-serving application writes sensitive user document fragments into broadly accessible telemetry events. Which implementation response is most appropriate for this ML engineering situation? Use supported safety filters and explicit application-level response validation rather than relying on prompts alone. Run continuous groundedness, citation relevance and task-completion evaluations on a held-out generative AI dataset. Restrict the workload service account to least-privilege roles and the minimum approved resource set. Redact or minimize sensitive prompt and response fields and restrict monitoring access with a documented retention policy. ExplanationCorrect Answer D:Observability data can itself expose confidential training or inference content if logs are collected without safeguards. To validate the correction: Inject synthetic secrets and verify redaction, log access denials and retention configuration. Other Answers A:Model instructions are probabilistic and safety guardrails need separate policy checks at input and output boundaries. It does not resolve the specific failure; the targeted correction is: Redact or minimize sensitive prompt and response fields and restrict monitoring access with a documented retention policy. B:Generative AI answer quality can regress through retrieval or prompt changes even when infrastructure is healthy. It does not resolve the specific failure; the targeted correction is: Redact or minimize sensitive prompt and response fields and restrict monitoring access with a documented retention policy. C:IAM controls cloud API permissions independently of whether a model produces accurate or safe predictions. It does not resolve the specific failure; the targeted correction is: Redact or minimize sensitive prompt and response fields and restrict monitoring access with a documented retention policy.
Question 1001A model-serving application writes sensitive user document fragments into broadly accessible telemetry events. Which technical mechanism is most relevant to this outcome? IAM controls cloud API permissions independently of whether a model produces accurate or safe predictions. Prompt injection is untrusted input attempting to override application instructions and can cause data leakage without isolation. Observability data can itself expose confidential training or inference content if logs are collected without safeguards. Training-serving skew can degrade predictions without any change to the underlying fitted model weights. ExplanationCorrect Answer C:Observability data can itself expose confidential training or inference content if logs are collected without safeguards. This supports the specific implementation: Redact or minimize sensitive prompt and response fields and restrict monitoring access with a documented retention policy. A suitable independent check is: Inject synthetic secrets and verify redaction, log access denials and retention configuration. Other Answers A:IAM controls cloud API permissions independently of whether a model produces accurate or safe predictions. The cause in the stated scenario is different: Observability data can itself expose confidential training or inference content if logs are collected without safeguards. B:Prompt injection is untrusted input attempting to override application instructions and can cause data leakage without isolation. The cause in the stated scenario is different: Observability data can itself expose confidential training or inference content if logs are collected without safeguards. D:Training-serving skew can degrade predictions without any change to the underlying fitted model weights. The cause in the stated scenario is different: Observability data can itself expose confidential training or inference content if logs are collected without safeguards.
Question 1002A model-serving application writes sensitive user document fragments into broadly accessible telemetry events. Which evidence would best demonstrate that this ML solution meets the requirement? Inject synthetic secrets and verify redaction, log access denials and retention configuration. Compare group metrics and representative false negatives on consented and legally appropriate audit samples. Apply a safe controlled load and verify p95 latency, error-rate and resource alarms trigger with clear ownership. Run allowed and denied access tests using the actual serving identity and audit role grants. ExplanationCorrect Answer A:Inject synthetic secrets and verify redaction, log access denials and retention configuration. This tests the underlying failure: Observability data can itself expose confidential training or inference content if logs are collected without safeguards. Other Answers B:Aggregate accuracy can obscure uneven subgroup errors and biased training labels or sampling distributions. That does not establish the required outcome; verify it by checking: Inject synthetic secrets and verify redaction, log access denials and retention configuration. C:Reliability metrics expose serving failures independently from predictive model quality and training drift. That does not establish the required outcome; verify it by checking: Inject synthetic secrets and verify redaction, log access denials and retention configuration. D:IAM controls cloud API permissions independently of whether a model produces accurate or safe predictions. That does not establish the required outcome; verify it by checking: Inject synthetic secrets and verify redaction, log access denials and retention configuration.
Question 1003A model-serving application writes sensitive user document fragments into broadly accessible telemetry events. What governance or operational safeguard should accompany the implementation? Apply data-classification and logging reviews before enabling verbose AI application diagnostics. Refresh reference distributions only after approved investigation and maintain source-drift incident runbooks. Version explainability methodology and require explanation quality reviews when features or architectures change. Maintain prompt-injection regression tests, least-privilege retrieval and incident review for new knowledge sources. ExplanationCorrect Answer A:Apply data-classification and logging reviews before enabling verbose AI application diagnostics. This protects the selected approach because Observability data can itself expose confidential training or inference content if logs are collected without safeguards. The immediate solution is: Redact or minimize sensitive prompt and response fields and restrict monitoring access with a documented retention policy. Other Answers B:Data drift is a change in input distributions and can occur even when prediction-outcome labels are delayed. The continuing safeguard required here is: Apply data-classification and logging reviews before enabling verbose AI application diagnostics. C:Feature attribution describes how inputs relate to predictions, but explanations are not necessarily causal proof. The continuing safeguard required here is: Apply data-classification and logging reviews before enabling verbose AI application diagnostics. D:Prompt injection is untrusted input attempting to override application instructions and can cause data leakage without isolation. The continuing safeguard required here is: Apply data-classification and logging reviews before enabling verbose AI application diagnostics.
Question 1004A support assistant stores private conversation content in debugging traces beyond the organization's retention policy. Which Google Cloud machine-learning design choice best satisfies the stated requirement? Redact or minimize sensitive prompt and response fields and restrict monitoring access with a documented retention policy. Monitor feature-attribution drift alongside outcome and input metrics on representative authorized samples. Use supported model-explainability methods on Agent Platform and validate attribution against domain expectations. Evaluate model outcomes and error rates across relevant groups and apply documented responsible-AI mitigations. ExplanationCorrect Answer A:Observability data can itself expose confidential training or inference content if logs are collected without safeguards. To validate the correction: Inject synthetic secrets and verify redaction, log access denials and retention configuration. Other Answers B:Attribution drift can signal changes in how a model uses information that overall aggregate metrics may miss. It does not resolve the specific failure; the targeted correction is: Redact or minimize sensitive prompt and response fields and restrict monitoring access with a documented retention policy. C:Feature attribution describes how inputs relate to predictions, but explanations are not necessarily causal proof. It does not resolve the specific failure; the targeted correction is: Redact or minimize sensitive prompt and response fields and restrict monitoring access with a documented retention policy. D:Aggregate accuracy can obscure uneven subgroup errors and biased training labels or sampling distributions. It does not resolve the specific failure; the targeted correction is: Redact or minimize sensitive prompt and response fields and restrict monitoring access with a documented retention policy.
Question 1005A support assistant stores private conversation content in debugging traces beyond the organization's retention policy. Which underlying ML principle makes the proposed approach effective? Prompt injection is untrusted input attempting to override application instructions and can cause data leakage without isolation. Model instructions are probabilistic and safety guardrails need separate policy checks at input and output boundaries. Training-serving skew can degrade predictions without any change to the underlying fitted model weights. Observability data can itself expose confidential training or inference content if logs are collected without safeguards. ExplanationCorrect Answer D:Observability data can itself expose confidential training or inference content if logs are collected without safeguards. This supports the specific implementation: Redact or minimize sensitive prompt and response fields and restrict monitoring access with a documented retention policy. A suitable independent check is: Inject synthetic secrets and verify redaction, log access denials and retention configuration. Other Answers A:Prompt injection is untrusted input attempting to override application instructions and can cause data leakage without isolation. The cause in the stated scenario is different: Observability data can itself expose confidential training or inference content if logs are collected without safeguards. B:Model instructions are probabilistic and safety guardrails need separate policy checks at input and output boundaries. The cause in the stated scenario is different: Observability data can itself expose confidential training or inference content if logs are collected without safeguards. C:Training-serving skew can degrade predictions without any change to the underlying fitted model weights. The cause in the stated scenario is different: Observability data can itself expose confidential training or inference content if logs are collected without safeguards.
Question 1006A support assistant stores private conversation content in debugging traces beyond the organization's retention policy. What controlled validation most directly confirms the approach worked? Run allowed and denied access tests using the actual serving identity and audit role grants. Replay shifted fixture data and confirm configured drift thresholds and alert routing work as expected. Inject synthetic secrets and verify redaction, log access denials and retention configuration. Submit defined unsafe test inputs and inspect filter decisions, false positives and user-safe fallback responses. ExplanationCorrect Answer C:Inject synthetic secrets and verify redaction, log access denials and retention configuration. This tests the underlying failure: Observability data can itself expose confidential training or inference content if logs are collected without safeguards. Other Answers A:IAM controls cloud API permissions independently of whether a model produces accurate or safe predictions. That does not establish the required outcome; verify it by checking: Inject synthetic secrets and verify redaction, log access denials and retention configuration. B:Data drift is a change in input distributions and can occur even when prediction-outcome labels are delayed. That does not establish the required outcome; verify it by checking: Inject synthetic secrets and verify redaction, log access denials and retention configuration. D:Model instructions are probabilistic and safety guardrails need separate policy checks at input and output boundaries. That does not establish the required outcome; verify it by checking: Inject synthetic secrets and verify redaction, log access denials and retention configuration.
Question 1007A support assistant stores private conversation content in debugging traces beyond the organization's retention policy. Which repeatable process control should the ML engineering team adopt? Require fairness monitoring, documented mitigation ownership and human escalation for high-impact decisions. Version retrieval indexes and evaluation prompts and gate releases on grounded-answer regression tests. Apply data-classification and logging reviews before enabling verbose AI application diagnostics. Require periodic IAM entitlement recertification and automatic policy checks before model deployment. ExplanationCorrect Answer C:Apply data-classification and logging reviews before enabling verbose AI application diagnostics. This protects the selected approach because Observability data can itself expose confidential training or inference content if logs are collected without safeguards. The immediate solution is: Redact or minimize sensitive prompt and response fields and restrict monitoring access with a documented retention policy. Other Answers A:Aggregate accuracy can obscure uneven subgroup errors and biased training labels or sampling distributions. The continuing safeguard required here is: Apply data-classification and logging reviews before enabling verbose AI application diagnostics. B:Generative AI answer quality can regress through retrieval or prompt changes even when infrastructure is healthy. The continuing safeguard required here is: Apply data-classification and logging reviews before enabling verbose AI application diagnostics. D:IAM controls cloud API permissions independently of whether a model produces accurate or safe predictions. The continuing safeguard required here is: Apply data-classification and logging reviews before enabling verbose AI application diagnostics.
Question 1008A production model endpoint receives requests using a service account with unnecessary organization-wide privileges. Which proposed approach best resolves the described model engineering problem? Use supported model-explainability methods on Agent Platform and validate attribution against domain expectations. Restrict the workload service account to least-privilege roles and the minimum approved resource set. Add training-serving feature-skew checks and deploy the exact versioned preprocessing artifacts used during training. Use supported safety filters and explicit application-level response validation rather than relying on prompts alone. ExplanationCorrect Answer B:IAM controls cloud API permissions independently of whether a model produces accurate or safe predictions. To validate the correction: Run allowed and denied access tests using the actual serving identity and audit role grants. Other Answers A:Feature attribution describes how inputs relate to predictions, but explanations are not necessarily causal proof. It does not resolve the specific failure; the targeted correction is: Restrict the workload service account to least-privilege roles and the minimum approved resource set. C:Training-serving skew can degrade predictions without any change to the underlying fitted model weights. It does not resolve the specific failure; the targeted correction is: Restrict the workload service account to least-privilege roles and the minimum approved resource set. D:Model instructions are probabilistic and safety guardrails need separate policy checks at input and output boundaries. It does not resolve the specific failure; the targeted correction is: Restrict the workload service account to least-privilege roles and the minimum approved resource set.
Question 1009A production model endpoint receives requests using a service account with unnecessary organization-wide privileges. Which technical mechanism is most relevant to this outcome? Prompt injection is untrusted input attempting to override application instructions and can cause data leakage without isolation. Aggregate accuracy can obscure uneven subgroup errors and biased training labels or sampling distributions. IAM controls cloud API permissions independently of whether a model produces accurate or safe predictions. Training-serving skew can degrade predictions without any change to the underlying fitted model weights. ExplanationCorrect Answer C:IAM controls cloud API permissions independently of whether a model produces accurate or safe predictions. This supports the specific implementation: Restrict the workload service account to least-privilege roles and the minimum approved resource set. A suitable independent check is: Run allowed and denied access tests using the actual serving identity and audit role grants. Other Answers A:Prompt injection is untrusted input attempting to override application instructions and can cause data leakage without isolation. The cause in the stated scenario is different: IAM controls cloud API permissions independently of whether a model produces accurate or safe predictions. B:Aggregate accuracy can obscure uneven subgroup errors and biased training labels or sampling distributions. The cause in the stated scenario is different: IAM controls cloud API permissions independently of whether a model produces accurate or safe predictions. D:Training-serving skew can degrade predictions without any change to the underlying fitted model weights. The cause in the stated scenario is different: IAM controls cloud API permissions independently of whether a model produces accurate or safe predictions.
Question 1010A production model endpoint receives requests using a service account with unnecessary organization-wide privileges. What is the most relevant acceptance check for the proposed ML change? Run allowed and denied access tests using the actual serving identity and audit role grants. Submit defined unsafe test inputs and inspect filter decisions, false positives and user-safe fallback responses. Apply a safe controlled load and verify p95 latency, error-rate and resource alarms trigger with clear ownership. Compare local attribution stability on similar cases and check against controlled perturbations and business rules. ExplanationCorrect Answer A:Run allowed and denied access tests using the actual serving identity and audit role grants. This tests the underlying failure: IAM controls cloud API permissions independently of whether a model produces accurate or safe predictions. Other Answers B:Model instructions are probabilistic and safety guardrails need separate policy checks at input and output boundaries. That does not establish the required outcome; verify it by checking: Run allowed and denied access tests using the actual serving identity and audit role grants. C:Reliability metrics expose serving failures independently from predictive model quality and training drift. That does not establish the required outcome; verify it by checking: Run allowed and denied access tests using the actual serving identity and audit role grants. D:Feature attribution describes how inputs relate to predictions, but explanations are not necessarily causal proof. That does not establish the required outcome; verify it by checking: Run allowed and denied access tests using the actual serving identity and audit role grants.