Question 280 A retrieved contract includes a paragraph instructing an agent to send the customer's credentials to a web endpoint. Which design change directly satisfies the stated constraints? Add explicit immutable-token preservation rules with input/output checks for commands, identifiers and measurements. Redact or suppress sensitive values before tracing and use approved metadata-only observability for protected fields. Apply scope guardrails requiring educational information, uncertainty and referral for clinical decisions, and avoid actionable patient-specific prescriptions. Ignore document-embedded commands, constrain tool calls to server-side permissions and redact credentials from the model context. ExplanationCorrect Answer D:The action targets the stated cause: Retrieved passages are untrusted task data, not higher-priority instructions authorizing external actions. Check the repair by: inject the malicious paragraph in a test corpus and confirm no network tool receives credential values. Other Answers A:Retrieved passages are untrusted task data, not higher-priority instructions authorizing external actions. B:Retrieved passages are untrusted task data, not higher-priority instructions authorizing external actions. C:Retrieved passages are untrusted task data, not higher-priority instructions authorizing external actions.
Question 281 A retrieved contract includes a paragraph instructing an agent to send the customer's credentials to a web endpoint. Which causal explanation best accounts for the failure? Output masking does not protect logs or tool traces; leakage can occur earlier in the data path. Language translation should transform prose while preserving tokens that are meaningful machine identifiers. Generative confidence is not clinical verification; missing patient context makes definitive treatment unsafe. Retrieved passages are untrusted task data, not higher-priority instructions authorizing external actions. ExplanationCorrect Answer D:Ignore document-embedded commands, constrain tool calls to server-side permissions and redact credentials from the model context. Other Answers A:That mechanism describes another problem: A support bot exposes the full credit-card number in MLflow traces while producing a redacted response to the customer. It does not explain this case, where retrieved passages are untrusted task data, not higher-priority instructions authorizing external actions. B:Retrieved passages are untrusted task data, not higher-priority instructions authorizing external actions. C:Retrieved passages are untrusted task data, not higher-priority instructions authorizing external actions.
Question 282 A retrieved contract includes a paragraph instructing an agent to send the customer's credentials to a web endpoint. Which concrete acceptance test best demonstrates the proposed change works? Run ambiguous clinical prompts and verify the system declines unsupported prescribing while providing appropriate general guidance. Inject the malicious paragraph in a test corpus and confirm no network tool receives credential values. Inject synthetic card data and verify neither trace attributes nor model outputs contain the raw identifier. Diff translated outputs against source commands and part numbers on a multilingual regression corpus. ExplanationCorrect Answer B:Retrieved passages are untrusted task data, not higher-priority instructions authorizing external actions. Other Answers A:Inject the malicious paragraph in a test corpus and confirm no network tool receives credential values. C:Inject the malicious paragraph in a test corpus and confirm no network tool receives credential values. D:Inject the malicious paragraph in a test corpus and confirm no network tool receives credential values.
Question 283 A retrieved contract includes a paragraph instructing an agent to send the customer's credentials to a web endpoint. Which operational practice should be made a durable requirement? Enforce least-privilege tools and continuous prompt-injection adversarial tests. Maintain protected-token extraction and post-translation equality checks for technical content. Apply end-to-end data minimization and log-scrubbing checks to all agent components. Maintain high-risk domain policies and independently review clinical safety evaluation cases. ExplanationCorrect Answer A:Retrieved passages are untrusted task data, not higher-priority instructions authorizing external actions. Other Answers B:That retrieved passages are untrusted task data, not higher-priority instructions authorizing external actions. C:Retrieved passages are untrusted task data, not higher-priority instructions authorizing external actions. D:This control addresses a different exposure: A medical Q&A assistant is asked for a diagnosis despite incomplete records, and it confidently suggests a specific prescription. It would not prevent this failure, because retrieved passages are untrusted task data, not higher-priority instructions authorizing external actions.
Question 284 A medical Q&A assistant is asked for a diagnosis despite incomplete records, and it confidently suggests a specific prescription. What should the implementation team change first to resolve the described defect? Require use of supplied specification fields, forbid unsupported values and include an explicit not_found response for missing attributes. Apply scope guardrails requiring educational information, uncertainty and referral for clinical decisions, and avoid actionable patient-specific prescriptions. Redact or suppress sensitive values before tracing and use approved metadata-only observability for protected fields. Ignore document-embedded commands, constrain tool calls to server-side permissions and redact credentials from the model context. ExplanationCorrect Answer B:The action targets the stated cause: Generative confidence is not clinical verification; missing patient context makes definitive treatment unsafe. Check the repair by: run ambiguous clinical prompts and verify the system declines unsupported prescribing while providing appropriate general guidance. Other Answers A:Generative confidence is not clinical verification; missing patient context makes definitive treatment unsafe. C:Generative confidence is not clinical verification; missing patient context makes definitive treatment unsafe. D:Generative confidence is not clinical verification; missing patient context makes definitive treatment unsafe.
Question 285 A medical Q&A assistant is asked for a diagnosis despite incomplete records, and it confidently suggests a specific prescription. Why does the current approach fail under these conditions? Retrieved passages are untrusted task data, not higher-priority instructions authorizing external actions. Output masking does not protect logs or tool traces; leakage can occur earlier in the data path. General prompts allow parametric guesses to compete with retrieved evidence; instruction hierarchy must make source-grounded facts decisive. Generative confidence is not clinical verification; missing patient context makes definitive treatment unsafe. ExplanationCorrect Answer D:Apply scope guardrails requiring educational information, uncertainty and referral for clinical decisions, and avoid actionable patient-specific prescriptions. Other Answers A:Generative confidence is not clinical verification; missing patient context makes definitive treatment unsafe. B:Generative confidence is not clinical verification; missing patient context makes definitive treatment unsafe. C:That mechanism describes another problem: An assistant repeatedly hallucinates product specifications when asked “tell me everything,” despite having retrieved product sheets. It does not explain this case, where generative confidence is not clinical verification; missing patient context makes definitive treatment unsafe.
Question 286 A medical Q&A assistant is asked for a diagnosis despite incomplete records, and it confidently suggests a specific prescription. Which observation would establish that the requirement is now met? Inject the malicious paragraph in a test corpus and confirm no network tool receives credential values. Run ambiguous clinical prompts and verify the system declines unsupported prescribing while providing appropriate general guidance. Test deliberately absent specifications and verify the answer says unavailable instead of supplying invented measurements. Inject synthetic card data and verify neither trace attributes nor model outputs contain the raw identifier. ExplanationCorrect Answer B:Generative confidence is not clinical verification; missing patient context makes definitive treatment unsafe. Other Answers A:Run ambiguous clinical prompts and verify the system declines unsupported prescribing while providing appropriate general guidance. C:Run ambiguous clinical prompts and verify the system declines unsupported prescribing while providing appropriate general guidance. D:Run ambiguous clinical prompts and verify the system declines unsupported prescribing while providing appropriate general guidance.
Question 287 A medical Q&A assistant is asked for a diagnosis despite incomplete records, and it confidently suggests a specific prescription. What lasting control best prevents recurrence of the identified issue? Apply end-to-end data minimization and log-scrubbing checks to all agent components. Measure groundedness on missing-field and conflicting-source examples during every prompt revision. Enforce least-privilege tools and continuous prompt-injection adversarial tests. Maintain high-risk domain policies and independently review clinical safety evaluation cases. ExplanationCorrect Answer D:Generative confidence is not clinical verification; missing patient context makes definitive treatment unsafe. Other Answers A:Generative confidence is not clinical verification; missing patient context makes definitive treatment unsafe. B:This control addresses a different exposure: An assistant repeatedly hallucinates product specifications when asked “tell me everything,” despite having retrieved product sheets. It would not prevent this failure, because generative confidence is not clinical verification; missing patient context makes definitive treatment unsafe. C:That generative confidence is not clinical verification; missing patient context makes definitive treatment unsafe.
Question 288 A support bot exposes the full credit-card number in MLflow traces while producing a redacted response to the customer. What is the most targeted corrective step for this observed failure? Redact or suppress sensitive values before tracing and use approved metadata-only observability for protected fields. Apply scope guardrails requiring educational information, uncertainty and referral for clinical decisions, and avoid actionable patient-specific prescriptions. Treat the benchmark as insufficient, run an Indonesian task evaluation and prefer a model with proven target-language behavior. Ignore document-embedded commands, constrain tool calls to server-side permissions and redact credentials from the model context. ExplanationCorrect Answer A:The action targets the stated cause: Output masking does not protect logs or tool traces; leakage can occur earlier in the data path. Check the repair by: inject synthetic card data and verify neither trace attributes nor model outputs contain the raw identifier. Other Answers B:Output masking does not protect logs or tool traces; leakage can occur earlier in the data path. C:Output masking does not protect logs or tool traces; leakage can occur earlier in the data path. D:Output masking does not protect logs or tool traces; leakage can occur earlier in the data path.
Question 289 A support bot exposes the full credit-card number in MLflow traces while producing a redacted response to the customer. What is the decisive mechanism behind the observed result? Output masking does not protect logs or tool traces; leakage can occur earlier in the data path. Model-card benchmark coverage does not automatically generalize to untested languages or task distributions. Generative confidence is not clinical verification; missing patient context makes definitive treatment unsafe. Retrieved passages are untrusted task data, not higher-priority instructions authorizing external actions. ExplanationCorrect Answer A:Redact or suppress sensitive values before tracing and use approved metadata-only observability for protected fields. Other Answers B:Output masking does not protect logs or tool traces; leakage can occur earlier in the data path. C:That mechanism describes another problem: A medical Q&A assistant is asked for a diagnosis despite incomplete records, and it confidently suggests a specific prescription. It does not explain this case, where output masking does not protect logs or tool traces; leakage can occur earlier in the data path. D:Output masking does not protect logs or tool traces; leakage can occur earlier in the data path.
Question 290 A support bot exposes the full credit-card number in MLflow traces while producing a redacted response to the customer. Which verification is strongest for this particular failure mode? Compare response correctness on a labeled Indonesian set and document coverage gaps from the model card. Inject the malicious paragraph in a test corpus and confirm no network tool receives credential values. Inject synthetic card data and verify neither trace attributes nor model outputs contain the raw identifier. Run ambiguous clinical prompts and verify the system declines unsupported prescribing while providing appropriate general guidance. ExplanationCorrect Answer C:Output masking does not protect logs or tool traces; leakage can occur earlier in the data path. Other Answers A:Inject synthetic card data and verify neither trace attributes nor model outputs contain the raw identifier. B:Inject synthetic card data and verify neither trace attributes nor model outputs contain the raw identifier. D:Inject synthetic card data and verify neither trace attributes nor model outputs contain the raw identifier.
