Question 20A document server serves requested files by joining a web root with unchecked URL segments; requests with dot-dot paths reveal configuration secrets. What should be institutionalized before future deployments? Require storage-root authorization checks after path normalization in every file endpoint. Add query-parameter tests and code review checks for all database interfaces. Combine CVSS with vulnerability intelligence, asset exposure and compensating controls in risk-based patch triage. Baseline permitted TLS versions and cryptographic policy in continuous configuration compliance tests. ExplanationCorrect Answer A:Require storage-root authorization checks after path normalization in every file endpoint. This safeguard limits repeat failures because path traversal abuses relative path elements to escape the intended directory, independent of file extension checks. The direct correction remains: Canonicalize paths and enforce that the resulting file remains inside the authorized storage root. Other Answers B:SQL injection occurs when untrusted input changes the syntax of a query rather than remaining a bound value. The lasting safeguard is to require storage-root authorization checks after path normalization in every file endpoint. C:CVSS summarizes severity but does not by itself account for live exploitation and local asset criticality. Reducing this risk requires a practice to require storage-root authorization checks after path normalization in every file endpoint. D:Encryption strength depends on negotiated algorithms and key handling; a TLS label alone provides no assurance. The stronger recurring control is to require storage-root authorization checks after path normalization in every file endpoint.
Question 21A security team prioritizes a high CVSS package flaw on an isolated test host over an actively exploited medium CVSS flaw on an internet-facing VPN gateway. What should the engineer change first to meet these conditions? Rank remediation by exploit availability, exposure and business impact alongside the published CVSS score. Replace string concatenation with parameterized queries and keep database privileges limited to required tables. Reconcile indicators with source updates and expire stale blocks using time-to-live and review controls. Canonicalize paths and enforce that the resulting file remains inside the authorized storage root. ExplanationCorrect Answer A:Rank remediation by exploit availability, exposure and business impact alongside the published CVSS score. This action specifically works because cVSS summarizes severity but does not by itself account for live exploitation and local asset criticality. Verify with: Check that the exploited exposed gateway receives an owner and remediation date before the isolated test host. Other Answers B:SQL injection occurs when untrusted input changes the syntax of a query rather than remaining a bound value. A more relevant action is to rank remediation by exploit availability, exposure and business impact alongside the published CVSS score. C:Indicators have limited useful lifetime; stale intelligence can create avoidable business impact. The required change instead is to rank remediation by exploit availability, exposure and business impact alongside the published CVSS score. D:Path traversal abuses relative path elements to escape the intended directory, independent of file extension checks. The required change instead is to rank remediation by exploit availability, exposure and business impact alongside the published CVSS score.
Question 22A security team prioritizes a high CVSS package flaw on an isolated test host over an actively exploited medium CVSS flaw on an internet-facing VPN gateway. Which cause-and-effect relationship determines the best answer? Path traversal abuses relative path elements to escape the intended directory, independent of file extension checks. Indicators have limited useful lifetime; stale intelligence can create avoidable business impact. CVSS summarizes severity but does not by itself account for live exploitation and local asset criticality. SQL injection occurs when untrusted input changes the syntax of a query rather than remaining a bound value. ExplanationCorrect Answer C:CVSS summarizes severity but does not by itself account for live exploitation and local asset criticality. The practical corrective action is: Rank remediation by exploit availability, exposure and business impact alongside the published CVSS score. Confirm it by: Check that the exploited exposed gateway receives an owner and remediation date before the isolated test host. Other Answers A:Path traversal abuses relative path elements to escape the intended directory, independent of file extension checks. In the stated system, CVSS summarizes severity but does not by itself account for live exploitation and local asset criticality. B:Indicators have limited useful lifetime; stale intelligence can create avoidable business impact. In the stated system, CVSS summarizes severity but does not by itself account for live exploitation and local asset criticality. D:SQL injection occurs when untrusted input changes the syntax of a query rather than remaining a bound value. By contrast, CVSS summarizes severity but does not by itself account for live exploitation and local asset criticality.
Question 23A security team prioritizes a high CVSS package flaw on an isolated test host over an actively exploited medium CVSS flaw on an internet-facing VPN gateway. Which observable result shows the correction is effective? Review indicator age, source reputation and a known-safe request after removing the obsolete block. Repeat the malicious payload in a negative test and confirm it returns no unauthorized records. Check that the exploited exposed gateway receives an owner and remediation date before the isolated test host. Test encoded and plain traversal attempts against allowed and forbidden paths after the change. ExplanationCorrect Answer C:Check that the exploited exposed gateway receives an owner and remediation date before the isolated test host. The evidence confirms the intended action: Rank remediation by exploit availability, exposure and business impact alongside the published CVSS score. It matters because cVSS summarizes severity but does not by itself account for live exploitation and local asset criticality. Other Answers A:Indicators have limited useful lifetime; stale intelligence can create avoidable business impact. The relevant acceptance check is to check that the exploited exposed gateway receives an owner and remediation date before the isolated test host. B:SQL injection occurs when untrusted input changes the syntax of a query rather than remaining a bound value. The failure is checked more directly by a test to check that the exploited exposed gateway receives an owner and remediation date before the isolated test host. D:Path traversal abuses relative path elements to escape the intended directory, independent of file extension checks. The relevant acceptance check is to check that the exploited exposed gateway receives an owner and remediation date before the isolated test host.
Question 24A security team prioritizes a high CVSS package flaw on an isolated test host over an actively exploited medium CVSS flaw on an internet-facing VPN gateway. What should be institutionalized before future deployments? Require storage-root authorization checks after path normalization in every file endpoint. Combine CVSS with vulnerability intelligence, asset exposure and compensating controls in risk-based patch triage. Add query-parameter tests and code review checks for all database interfaces. Automate IOC lifecycle management with owner, confidence, expiration and exception review. ExplanationCorrect Answer B:Combine CVSS with vulnerability intelligence, asset exposure and compensating controls in risk-based patch triage. This safeguard limits repeat failures because cVSS summarizes severity but does not by itself account for live exploitation and local asset criticality. The direct correction remains: Rank remediation by exploit availability, exposure and business impact alongside the published CVSS score. Other Answers A:Path traversal abuses relative path elements to escape the intended directory, independent of file extension checks. The lasting safeguard is to combine CVSS with vulnerability intelligence, asset exposure and compensating controls in risk-based patch triage. C:SQL injection occurs when untrusted input changes the syntax of a query rather than remaining a bound value. The stronger recurring control is to combine CVSS with vulnerability intelligence, asset exposure and compensating controls in risk-based patch triage. D:Indicators have limited useful lifetime; stale intelligence can create avoidable business impact. The lasting safeguard is to combine CVSS with vulnerability intelligence, asset exposure and compensating controls in risk-based patch triage.
Question 25A retrieval assistant reads a customer-supplied PDF containing instructions to ignore the security policy and disclose another customer's report. What should the engineer change first to meet these conditions? Pin and verify signed package provenance and isolate the deployment until the dependency is investigated. Treat retrieved text as untrusted data, block tool actions from document instructions and enforce tenant authorization independently. Replace string concatenation with parameterized queries and keep database privileges limited to required tables. Remove or quarantine the untrusted embedding entry and require source authority filters before retrieval results reach generation. ExplanationCorrect Answer B:Treat retrieved text as untrusted data, block tool actions from document instructions and enforce tenant authorization independently. This action specifically works because prompt injection crosses a trust boundary when lower-trust source content is promoted into privileged instructions. Verify with: Replay the malicious PDF and confirm no cross-tenant lookup or forbidden tool call occurs. Other Answers A:An LLM supply-chain component can execute code or alter prompts outside the model's ordinary content controls. The corrective action for this requirement is to treat retrieved text as untrusted data, block tool actions from document instructions and enforce tenant authorization independently. C:SQL injection occurs when untrusted input changes the syntax of a query rather than remaining a bound value. The corrective action for this requirement is to treat retrieved text as untrusted data, block tool actions from document instructions and enforce tenant authorization independently. D:Vector similarity establishes semantic proximity, not the provenance or trustworthiness of retrieved documents. This failure calls for a decision to treat retrieved text as untrusted data, block tool actions from document instructions and enforce tenant authorization independently.
Question 26A retrieval assistant reads a customer-supplied PDF containing instructions to ignore the security policy and disclose another customer's report. Which cause-and-effect relationship determines the best answer? Prompt injection crosses a trust boundary when lower-trust source content is promoted into privileged instructions. Vector similarity establishes semantic proximity, not the provenance or trustworthiness of retrieved documents. An LLM supply-chain component can execute code or alter prompts outside the model's ordinary content controls. SQL injection occurs when untrusted input changes the syntax of a query rather than remaining a bound value. ExplanationCorrect Answer A:Prompt injection crosses a trust boundary when lower-trust source content is promoted into privileged instructions. The practical corrective action is: Treat retrieved text as untrusted data, block tool actions from document instructions and enforce tenant authorization independently. Confirm it by: Replay the malicious PDF and confirm no cross-tenant lookup or forbidden tool call occurs. Other Answers B:Vector similarity establishes semantic proximity, not the provenance or trustworthiness of retrieved documents. In the stated system, prompt injection crosses a trust boundary when lower-trust source content is promoted into privileged instructions. C:An LLM supply-chain component can execute code or alter prompts outside the model's ordinary content controls. In the stated system, prompt injection crosses a trust boundary when lower-trust source content is promoted into privileged instructions. D:SQL injection occurs when untrusted input changes the syntax of a query rather than remaining a bound value. In the stated system, prompt injection crosses a trust boundary when lower-trust source content is promoted into privileged instructions.
Question 27A retrieval assistant reads a customer-supplied PDF containing instructions to ignore the security policy and disclose another customer's report. Which observable result shows the correction is effective? Replay the malicious PDF and confirm no cross-tenant lookup or forbidden tool call occurs. Compare SBOM and artifact digests to approved releases and inspect outbound connection telemetry. Requery affected topics and verify that citations resolve only to approved signed knowledge sources. Repeat the malicious payload in a negative test and confirm it returns no unauthorized records. ExplanationCorrect Answer A:Replay the malicious PDF and confirm no cross-tenant lookup or forbidden tool call occurs. The evidence confirms the intended action: Treat retrieved text as untrusted data, block tool actions from document instructions and enforce tenant authorization independently. It matters because prompt injection crosses a trust boundary when lower-trust source content is promoted into privileged instructions. Other Answers B:An LLM supply-chain component can execute code or alter prompts outside the model's ordinary content controls. A more meaningful verification is to replay the malicious PDF and confirm no cross-tenant lookup or forbidden tool call occurs. C:Vector similarity establishes semantic proximity, not the provenance or trustworthiness of retrieved documents. The relevant acceptance check is to replay the malicious PDF and confirm no cross-tenant lookup or forbidden tool call occurs. D:SQL injection occurs when untrusted input changes the syntax of a query rather than remaining a bound value. A more meaningful verification is to replay the malicious PDF and confirm no cross-tenant lookup or forbidden tool call occurs.
Question 28A retrieval assistant reads a customer-supplied PDF containing instructions to ignore the security policy and disclose another customer's report. What should be institutionalized before future deployments? Enforce signed dependency admission and scan model-serving containers and extensions before deployment. Add query-parameter tests and code review checks for all database interfaces. Audit embedding ingestion and apply provenance, freshness and owner approval gates to search indexes. Separate instruction layers from retrieved context and test adversarial document content in every release. ExplanationCorrect Answer D:Separate instruction layers from retrieved context and test adversarial document content in every release. This safeguard limits repeat failures because prompt injection crosses a trust boundary when lower-trust source content is promoted into privileged instructions. The direct correction remains: Treat retrieved text as untrusted data, block tool actions from document instructions and enforce tenant authorization independently. Other Answers A:An LLM supply-chain component can execute code or alter prompts outside the model's ordinary content controls. A more direct preventive measure is to separate instruction layers from retrieved context and test adversarial document content in every release. B:SQL injection occurs when untrusted input changes the syntax of a query rather than remaining a bound value. A more direct preventive measure is to separate instruction layers from retrieved context and test adversarial document content in every release. C:Vector similarity establishes semantic proximity, not the provenance or trustworthiness of retrieved documents. The lasting safeguard is to separate instruction layers from retrieved context and test adversarial document content in every release.
Question 29A support chatbot retrieves a poisoned knowledge snippet ranking a fake reset URL above the signed company recovery page. What should the engineer change first to meet these conditions? Pin and verify signed package provenance and isolate the deployment until the dependency is investigated. Treat retrieved text as untrusted data, block tool actions from document instructions and enforce tenant authorization independently. Follow the API pagination cursor or next link until the intended time range is completely retrieved. Remove or quarantine the untrusted embedding entry and require source authority filters before retrieval results reach generation. ExplanationCorrect Answer D:Remove or quarantine the untrusted embedding entry and require source authority filters before retrieval results reach generation. This action specifically works because vector similarity establishes semantic proximity, not the provenance or trustworthiness of retrieved documents. Verify with: Requery affected topics and verify that citations resolve only to approved signed knowledge sources. Other Answers A:An LLM supply-chain component can execute code or alter prompts outside the model's ordinary content controls. The required change instead is to remove or quarantine the untrusted embedding entry and require source authority filters before retrieval results reach generation. B:Prompt injection crosses a trust boundary when lower-trust source content is promoted into privileged instructions. The required change instead is to remove or quarantine the untrusted embedding entry and require source authority filters before retrieval results reach generation. C:REST API listing results can be paginated, so one successful HTTP 200 response may be only a partial dataset. The corrective action for this requirement is to remove or quarantine the untrusted embedding entry and require source authority filters before retrieval results reach generation.
Question 30A support chatbot retrieves a poisoned knowledge snippet ranking a fake reset URL above the signed company recovery page. Which cause-and-effect relationship determines the best answer? Prompt injection crosses a trust boundary when lower-trust source content is promoted into privileged instructions. REST API listing results can be paginated, so one successful HTTP 200 response may be only a partial dataset. An LLM supply-chain component can execute code or alter prompts outside the model's ordinary content controls. Vector similarity establishes semantic proximity, not the provenance or trustworthiness of retrieved documents. ExplanationCorrect Answer D:Vector similarity establishes semantic proximity, not the provenance or trustworthiness of retrieved documents. The practical corrective action is: Remove or quarantine the untrusted embedding entry and require source authority filters before retrieval results reach generation. Confirm it by: Requery affected topics and verify that citations resolve only to approved signed knowledge sources. Other Answers A:Prompt injection crosses a trust boundary when lower-trust source content is promoted into privileged instructions. By contrast, vector similarity establishes semantic proximity, not the provenance or trustworthiness of retrieved documents. B:REST API listing results can be paginated, so one successful HTTP 200 response may be only a partial dataset. In the stated system, vector similarity establishes semantic proximity, not the provenance or trustworthiness of retrieved documents. C:An LLM supply-chain component can execute code or alter prompts outside the model's ordinary content controls. By contrast, vector similarity establishes semantic proximity, not the provenance or trustworthiness of retrieved documents.