McAfee-Secured Website

Exam Bundle

Exam Code: CAS-005

Exam Name CompTIA SecurityX

Certification Provider: CompTIA

Corresponding Certification: CASP

CompTIA CAS-005 Bundle $44.99

100% CompTIA CAS-005 Guaranteed Success With Testking Training Materials

Get Real CAS-005 Exam Questions & Expert Verified Answers as Experienced on the Real Exam!

  • Questions & Answers

    CAS-005 Questions & Answers

    482 Questions & Answers

    The ultimate exam preparation tool, Questions & Answers mirrors the actual CAS-005 exam while allowing you to hone and measure every aspect of your test-taking ability.

  • CAS-005 Video Course

    CAS-005 Video Course

    180 Video Lectures

    CAS-005 Video Course is developed by CompTIA Professionals to help you pass the CAS-005 exam.

    Description

    This course will improve your knowledge and skills required to pass CompTIA SecurityX exam.
  • Study Guide

    CAS-005 Study Guide

    903 PDF Pages

    Developed by industry experts, this 903-page guide spells out in painstaking detail all of the information you need to ace CAS-005 exam.

CompTIA SecurityX (CAS-005) Practice Tests

Satisfaction Guaranteed

Money Back Guarantee

Testking provides hassle-free money back guarantee with our products. That is because we have 100% trust in the abilities of our professional and experience product team, and our record is a proof of that.

99.6% PASS RATE
Total Cost: $194.97
Bundle Price: $149.98

Purchase Individually

  • Questions & Answers

    Questions & Answers

    482 Questions

    $124.99
  • CAS-005 Video Course

    Video Course

    180 Video Lectures

    $39.99
  • Study Guide

    Study Guide

    903 PDF Pages

    $29.99

CompTIA CAS-005 Practice Test 11

Question 100

A follow-up assurance review considers this problem: A support chatbot can submit payment refunds using the employee's broad API token without approval thresholds. Which outcome best shows the intended improvement?

  1. A simulated convincing synthetic call cannot authorize transfer without independent second-channel confirmation.
  2. A simulated high-value refund is halted for approval while authorized low-value requests succeed.
  3. A test suite records controlled failure rates, safeguards, and remediation owners across attack categories.
  4. Contract administrators correctly reject a simulated payment-change request lacking dual approval.

Explanation

Correct Answer

B:

This is direct test evidence of the control objective in the case: A support chatbot can submit payment refunds using the employee's broad API token without approval thresholds. It demonstrates that the proposed response, restrict tool permissions to approved refund ranges and require independent authorization for high-value operations., produces the measurable result a simulated high-value refund is halted for approval while authorized low-value requests succeed. The observation tests a relevant security outcome rather than an unrelated control or reporting activity.

Incorrect Answers

A:

In “A marketing team uses synthetic executive audio to authorize an urgent bank transfer through a voice call”, “A simulated convincing synthetic call cannot authorize transfer without independent second-channel confirmation” is the acceptance signal for “Implement out-of-band verification and dual approval for sensitive transfer instructions regardless of voice similarity”. The required proof for the selected response “Restrict tool permissions to approved refund ranges and require independent authorization for high-value operations” is “A simulated high-value refund is halted for approval while authorized low-value requests succeed”.

C:

In the problem “A support chatbot can submit payment refunds using the employee's broad API token without approval thresholds”, evidence “A test suite records controlled failure rates, safeguards, and remediation owners across attack categories.” is not decisive because it tests “A team tests an AI model using only ordinary prompts, omitting model denial-of-service and extraction abuse cases”; the requested validation is a simulated high-value refund is halted for approval while authorized low-value requests succeed.

D:

For the issue “A support chatbot can submit payment refunds using the employee's broad API token without approval thresholds”, measuring “Contract administrators correctly reject a simulated payment-change request lacking dual approval.” would validate “A multinational's annual awareness program ignores contract administrators with access to supplier payment details” instead of the required control effect: a simulated high-value refund is halted for approval while authorized low-value requests succeed.

Question 101

A senior engineer examines the following finding: A retrieval-augmented assistant treats instructions hidden in supplier PDF content as higher-priority system commands. Which action most directly addresses the requirement?

  1. Restrict tool permissions to approved refund ranges and require independent authorization for high-value operations.
  2. Implement out-of-band verification and dual approval for sensitive transfer instructions regardless of voice similarity.
  3. Separate retrieved documents from trusted instructions and validate tool requests against independent policy.
  4. Compare prevention, detection, containment, and recovery layers and record remaining exposure against risk appetite.

Explanation

Correct Answer

C:

Separate retrieved documents from trusted instructions and validate tool requests against independent policy. The acceptance check is that malicious document instructions fail to trigger privileged tool calls in a controlled evaluation. That outcome checks whether the requirement is met, rather than treating completion of a security task as proof of effectiveness.

Incorrect Answers

A:

“A simulated high-value refund is halted for approval while authorized low-value requests succeed” would be meaningful evidence for “Restrict tool permissions to approved refund ranges and require independent authorization for high-value operations” in the case “A support chatbot can submit payment refunds using the employee's broad API token without approval thresholds”. It does not replace “Separate retrieved documents from trusted instructions and validate tool requests against independent policy” to correct “A retrieval-augmented assistant treats instructions hidden in supplier PDF content as higher-priority system commands” and verify “Malicious document instructions fail to trigger privileged tool calls in a controlled evaluation”.

B:

Although “Implement out-of-band verification and dual approval for sensitive transfer instructions regardless of voice similarity.” is a reasonable response to “A marketing team uses synthetic executive audio to authorize an urgent bank transfer through a voice call”, this case instead concerns “A retrieval-augmented assistant treats instructions hidden in supplier PDF content as higher-priority system commands”, for which success requires that malicious document instructions fail to trigger privileged tool calls in a controlled evaluation.

D:

This choice would be evaluated by “A threat exercise demonstrates controls for each phase and shows an accepted residual-risk decision” when dealing with “An enterprise faces this security constraint: A ransomware risk review says all consequences can be eliminated by procuring more endpoint licenses”. It does not deliver the distinct control “Separate retrieved documents from trusted instructions and validate tool requests against independent policy” needed for “A retrieval-augmented assistant treats instructions hidden in supplier PDF content as higher-priority system commands”.

Question 102

After the team responds to the following issue, which finding best demonstrates the desired security outcome? A retrieval-augmented assistant treats instructions hidden in supplier PDF content as higher-priority system commands.

  1. A release candidate has verifiable model hashes, supplier evidence, and traceable evaluation datasets.
  2. A test suite records controlled failure rates, safeguards, and remediation owners across attack categories.
  3. Malicious document instructions fail to trigger privileged tool calls in a controlled evaluation.
  4. A controlled data-subject request is processed under applicable documented requirements with a complete audit trail.

Explanation

Correct Answer

C:

This is direct test evidence of the control objective in the case: A retrieval-augmented assistant treats instructions hidden in supplier PDF content as higher-priority system commands. It demonstrates that the proposed response, separate retrieved documents from trusted instructions and validate tool requests against independent policy., produces the measurable result malicious document instructions fail to trigger privileged tool calls in a controlled evaluation. The observation tests a relevant security outcome rather than an unrelated control or reporting activity.

Incorrect Answers

A:

In the problem “A retrieval-augmented assistant treats instructions hidden in supplier PDF content as higher-priority system commands”, evidence “A release candidate has verifiable model hashes, supplier evidence, and traceable evaluation datasets.” is not decisive because it tests “A healthcare AI model's external training supplier cannot identify where checkpoint weights and evaluation data were sourced”; the requested validation is malicious document instructions fail to trigger privileged tool calls in a controlled evaluation.

B:

For the issue “A retrieval-augmented assistant treats instructions hidden in supplier PDF content as higher-priority system commands”, measuring “A test suite records controlled failure rates, safeguards, and remediation owners across attack categories.” would validate “A team tests an AI model using only ordinary prompts, omitting model denial-of-service and extraction abuse cases” instead of the required control effect: malicious document instructions fail to trigger privileged tool calls in a controlled evaluation.

D:

Observation “A controlled data-subject request is processed under applicable documented requirements with a complete audit trail.” would be useful for “A company deploys biometric recognition across borders without considering local data-subject access or correction rights”, but it does not prove that the response to “A retrieval-augmented assistant treats instructions hidden in supplier PDF content as higher-priority system commands” achieved the relevant outcome, malicious document instructions fail to trigger privileged tool calls in a controlled evaluation.

Question 103

An enterprise faces this security constraint: A model-training pipeline includes customer identifiers that are not necessary for the intended forecasting task. Which proposal best addresses the stated risk?

  1. Engage privacy counsel to assess applicable child-data consent, processing, retention, and contracting obligations.
  2. Apply data minimization, documented processing purpose, and privacy review before training.
  3. Restrict tool permissions to approved refund ranges and require independent authorization for high-value operations.
  4. Require explainability, human review, and source-traceable supporting indicators for consequential decisions.

Explanation

Correct Answer

B:

Apply data minimization, documented processing purpose, and privacy review before training. The acceptance check is that training inputs exclude unnecessary identifiers and a privacy audit traces approved data sources. That outcome checks whether the requirement is met, rather than treating completion of a security task as proof of effectiveness.

Incorrect Answers

A:

The option “Engage privacy counsel to assess applicable child-data consent, processing, retention, and contracting obligations.” assumes an issue like “A new subcontractor handles children's data, but procurement requests only a standard security questionnaire”, not the specific observed weakness “A model-training pipeline includes customer identifiers that are not necessary for the intended forecasting task”; it leaves the evidence goal unmet: training inputs exclude unnecessary identifiers and a privacy audit traces approved data sources.

C:

This choice would be evaluated by “A simulated high-value refund is halted for approval while authorized low-value requests succeed” when dealing with “A support chatbot can submit payment refunds using the employee's broad API token without approval thresholds”. It does not deliver the distinct control “Apply data minimization, documented processing purpose, and privacy review before training” needed for “An enterprise faces this security constraint: A model-training pipeline includes customer identifiers that are not necessary for the intended forecasting task”.

D:

Although “Require explainability, human review, and source-traceable supporting indicators for consequential decisions.” is a reasonable response to “An executive uses a generative AI risk score to deny supplier access without understanding confidence or evidence”, this case instead concerns “A model-training pipeline includes customer identifiers that are not necessary for the intended forecasting task”, for which success requires that training inputs exclude unnecessary identifiers and a privacy audit traces approved data sources.

Question 104

Which test or documented result best shows that the underlying issue was properly addressed? A model-training pipeline includes customer identifiers that are not necessary for the intended forecasting task.

  1. Random sampled interactions include the required AI-use notice and audit records.
  2. A test of the direct endpoint is denied while approved proxy traffic meets authorization policy.
  3. A simulated convincing synthetic call cannot authorize transfer without independent second-channel confirmation.
  4. Training inputs exclude unnecessary identifiers and a privacy audit traces approved data sources.

Explanation

Correct Answer

D:

This is direct test evidence of the control objective in the case: A model-training pipeline includes customer identifiers that are not necessary for the intended forecasting task. It demonstrates that the proposed response, apply data minimization, documented processing purpose, and privacy review before training., produces the measurable result training inputs exclude unnecessary identifiers and a privacy audit traces approved data sources. The observation tests a relevant security outcome rather than an unrelated control or reporting activity.

Incorrect Answers

A:

The observed “Random sampled interactions include the required AI-use notice and audit records” validates the separate control “Define disclosure, retention, and oversight requirements for AI-mediated customer communications” under “A regulator asks whether customer support calls were AI-generated, but the organization has no disclosure policy”. The present decision calls for the different result “Training inputs exclude unnecessary identifiers and a privacy audit traces approved data sources” from “Apply data minimization, documented processing purpose, and privacy review before training”.

B:

Observation “A test of the direct endpoint is denied while approved proxy traffic meets authorization policy.” would be useful for “An enterprise threat model assumes a reverse proxy protects internal APIs but direct service addresses remain reachable”, but it does not prove that the response to “A model-training pipeline includes customer identifiers that are not necessary for the intended forecasting task” achieved the relevant outcome, training inputs exclude unnecessary identifiers and a privacy audit traces approved data sources.

C:

In the problem “A model-training pipeline includes customer identifiers that are not necessary for the intended forecasting task”, evidence “A simulated convincing synthetic call cannot authorize transfer without independent second-channel confirmation.” is not decisive because it tests “A marketing team uses synthetic executive audio to authorize an urgent bank transfer through a voice call”; the requested validation is training inputs exclude unnecessary identifiers and a privacy audit traces approved data sources.

Question 105

An enterprise security review reports: An executive uses a generative AI risk score to deny supplier access without understanding confidence or evidence. Which response best addresses the underlying security issue?

  1. Define an emergency change path with immediate authorized safeguards and compulsory post-implementation security review.
  2. Restrict tool permissions to approved refund ranges and require independent authorization for high-value operations.
  3. Require supplier provenance attestations, artifact integrity, and test-data chain of custody for model release.
  4. Require explainability, human review, and source-traceable supporting indicators for consequential decisions.

Explanation

Correct Answer

D:

Require explainability, human review, and source-traceable supporting indicators for consequential decisions. The acceptance check is that a sample decision includes verified underlying evidence, reviewer signoff, and an appeals process. That outcome checks whether the requirement is met, rather than treating completion of a security task as proof of effectiveness.

Incorrect Answers

A:

“Define an emergency change path with immediate authorized safeguards and compulsory post-implementation security review” addresses “An enterprise faces this security constraint: An operations team bypasses security review for emergency configuration changes and never performs retrospective validation”, with “Emergency tickets show time-bound approval, tested rollback, and a completed post-change risk assessment” as its acceptance test. Here the specific problem is “An executive uses a generative AI risk score to deny supplier access without understanding confidence or evidence”: it requires “Require explainability, human review, and source-traceable supporting indicators for consequential decisions” and evidence that “A sample decision includes verified underlying evidence, reviewer signoff, and an appeals process”.

B:

Although “Restrict tool permissions to approved refund ranges and require independent authorization for high-value operations.” is a reasonable response to “A support chatbot can submit payment refunds using the employee's broad API token without approval thresholds”, this case instead concerns “An executive uses a generative AI risk score to deny supplier access without understanding confidence or evidence”, for which success requires that a sample decision includes verified underlying evidence, reviewer signoff, and an appeals process.

C:

The option “Require supplier provenance attestations, artifact integrity, and test-data chain of custody for model release.” assumes an issue like “A healthcare AI model's external training supplier cannot identify where checkpoint weights and evaluation data were sourced”, not the specific observed weakness “An executive uses a generative AI risk score to deny supplier access without understanding confidence or evidence”; it leaves the evidence goal unmet: a sample decision includes verified underlying evidence, reviewer signoff, and an appeals process.

Question 106

A follow-up assurance review considers this problem: An executive uses a generative AI risk score to deny supplier access without understanding confidence or evidence. Which outcome best shows the intended improvement?

  1. Sample exceptions contain current owners, evidence, approval history, and active expiry reminders.
  2. A sample decision includes verified underlying evidence, reviewer signoff, and an appeals process.
  3. A release candidate has verifiable model hashes, supplier evidence, and traceable evaluation datasets.
  4. Malicious document instructions fail to trigger privileged tool calls in a controlled evaluation.

Explanation

Correct Answer

B:

This is direct test evidence of the control objective in the case: An executive uses a generative AI risk score to deny supplier access without understanding confidence or evidence. It demonstrates that the proposed response, require explainability, human review, and source-traceable supporting indicators for consequential decisions., produces the measurable result a sample decision includes verified underlying evidence, reviewer signoff, and an appeals process. The observation tests a relevant security outcome rather than an unrelated control or reporting activity.

Incorrect Answers

A:

In the problem “An executive uses a generative AI risk score to deny supplier access without understanding confidence or evidence”, evidence “Sample exceptions contain current owners, evidence, approval history, and active expiry reminders.” is not decisive because it tests “A cloud team creates separate exception spreadsheets that omit approval dates and residual-risk ownership”; the requested validation is a sample decision includes verified underlying evidence, reviewer signoff, and an appeals process.

C:

In “A healthcare AI model's external training supplier cannot identify where checkpoint weights and evaluation data were sourced”, “A release candidate has verifiable model hashes, supplier evidence, and traceable evaluation datasets” is the acceptance signal for “Require supplier provenance attestations, artifact integrity, and test-data chain of custody for model release”. The required proof for the selected response “Require explainability, human review, and source-traceable supporting indicators for consequential decisions” is “A sample decision includes verified underlying evidence, reviewer signoff, and an appeals process”.

D:

Observation “Malicious document instructions fail to trigger privileged tool calls in a controlled evaluation.” would be useful for “A retrieval-augmented assistant treats instructions hidden in supplier PDF content as higher-priority system commands”, but it does not prove that the response to “An executive uses a generative AI risk score to deny supplier access without understanding confidence or evidence” achieved the relevant outcome, a sample decision includes verified underlying evidence, reviewer signoff, and an appeals process.

Question 107

During an authorized security assessment, the team discovers: A marketing team uses synthetic executive audio to authorize an urgent bank transfer through a voice call. What should be prioritized?

  1. Implement out-of-band verification and dual approval for sensitive transfer instructions regardless of voice similarity.
  2. Add adversarial evaluation for rate exhaustion, sensitive output, prompt injection, and model extraction.
  3. Constrain agents to allowlisted actions and require approval for destructive or privileged execution.
  4. Evaluate data classification, cross-border transfer rules, contractual restrictions, and required approval before moving access.

Explanation

Correct Answer

A:

Implement out-of-band verification and dual approval for sensitive transfer instructions regardless of voice similarity. The acceptance check is that a simulated convincing synthetic call cannot authorize transfer without independent second-channel confirmation. That outcome checks whether the requirement is met, rather than treating completion of a security task as proof of effectiveness.

Incorrect Answers

B:

“Add adversarial evaluation for rate exhaustion, sensitive output, prompt injection, and model extraction” addresses “A team tests an AI model using only ordinary prompts, omitting model denial-of-service and extraction abuse cases”, with “A test suite records controlled failure rates, safeguards, and remediation owners across attack categories” as its acceptance test. Here the specific problem is “A marketing team uses synthetic executive audio to authorize an urgent bank transfer through a voice call”: it requires “Implement out-of-band verification and dual approval for sensitive transfer instructions regardless of voice similarity” and evidence that “A simulated convincing synthetic call cannot authorize transfer without independent second-channel confirmation”.

C:

“Constrain agents to allowlisted actions and require approval for destructive or privileged execution” addresses “An enterprise faces this security constraint: A virtual assistant executes shell commands derived from untrusted ticket descriptions without human authorization”, with “An injected ticket command is rejected by tool policy before reaching a shell” as its acceptance test. Here the specific problem is “A marketing team uses synthetic executive audio to authorize an urgent bank transfer through a voice call”: it requires “Implement out-of-band verification and dual approval for sensitive transfer instructions regardless of voice similarity” and evidence that “A simulated convincing synthetic call cannot authorize transfer without independent second-channel confirmation”.

D:

Although “Evaluate data classification, cross-border transfer rules, contractual restrictions, and required approval before moving access.” is a reasonable response to “An international supplier proposes to move restricted research telemetry to a foreign support location without export review”, this case instead concerns “A marketing team uses synthetic executive audio to authorize an urgent bank transfer through a voice call”, for which success requires that a simulated convincing synthetic call cannot authorize transfer without independent second-channel confirmation.

Question 108

After the team responds to the following issue, which finding best demonstrates the desired security outcome? A marketing team uses synthetic executive audio to authorize an urgent bank transfer through a voice call.

  1. A simulated high-value refund is halted for approval while authorized low-value requests succeed.
  2. A release candidate has verifiable model hashes, supplier evidence, and traceable evaluation datasets.
  3. A simulated convincing synthetic call cannot authorize transfer without independent second-channel confirmation.
  4. A test compromise of production credentials cannot delete recovery copies and an isolated restore succeeds.

Explanation

Correct Answer

C:

This is direct test evidence of the control objective in the case: A marketing team uses synthetic executive audio to authorize an urgent bank transfer through a voice call. It demonstrates that the proposed response, implement out-of-band verification and dual approval for sensitive transfer instructions regardless of voice similarity., produces the measurable result a simulated convincing synthetic call cannot authorize transfer without independent second-channel confirmation. The observation tests a relevant security outcome rather than an unrelated control or reporting activity.

Incorrect Answers

A:

“A simulated high-value refund is halted for approval while authorized low-value requests succeed” would verify “Restrict tool permissions to approved refund ranges and require independent authorization for high-value operations” for the situation “A support chatbot can submit payment refunds using the employee's broad API token without approval thresholds”. Here the required evidence is “A simulated convincing synthetic call cannot authorize transfer without independent second-channel confirmation” after “Implement out-of-band verification and dual approval for sensitive transfer instructions regardless of voice similarity”; the alternative observation does not test that control.

B:

Observation “A release candidate has verifiable model hashes, supplier evidence, and traceable evaluation datasets.” would be useful for “A healthcare AI model's external training supplier cannot identify where checkpoint weights and evaluation data were sourced”, but it does not prove that the response to “A marketing team uses synthetic executive audio to authorize an urgent bank transfer through a voice call” achieved the relevant outcome, a simulated convincing synthetic call cannot authorize transfer without independent second-channel confirmation.

D:

For the issue “A marketing team uses synthetic executive audio to authorize an urgent bank transfer through a voice call”, measuring “A test compromise of production credentials cannot delete recovery copies and an isolated restore succeeds.” would validate “A hospital backup service shares administrators and credentials with production ransomware targets” instead of the required control effect: a simulated convincing synthetic call cannot authorize transfer without independent second-channel confirmation.

Question 109

A cross-functional security group is considering a decision: An internal coding assistant sends proprietary source files to a public model endpoint despite a restricted-data policy. What is the most appropriate control or investigation response?

  1. Separate backup control-plane identity, enforce immutable offline recovery copies, and validate isolated restoration.
  2. Enforce approved model endpoints, source-code classification checks, and egress control on AI integrations.
  3. Restrict tool permissions to approved refund ranges and require independent authorization for high-value operations.
  4. Add adversarial evaluation for rate exhaustion, sensitive output, prompt injection, and model extraction.

Explanation

Correct Answer

B:

Enforce approved model endpoints, source-code classification checks, and egress control on AI integrations. The acceptance check is that test uploads of restricted repositories are blocked and logged without transferring code externally. That outcome checks whether the requirement is met, rather than treating completion of a security task as proof of effectiveness.

Incorrect Answers

A:

The proposal “Separate backup control-plane identity, enforce immutable offline recovery copies, and validate isolated restoration” manages “A hospital backup service shares administrators and credentials with production ransomware targets”, where successful operation means “A test compromise of production credentials cannot delete recovery copies and an isolated restore succeeds”. The present issue instead needs “Enforce approved model endpoints, source-code classification checks, and egress control on AI integrations” to establish “Test uploads of restricted repositories are blocked and logged without transferring code externally”.

C:

The option “Restrict tool permissions to approved refund ranges and require independent authorization for high-value operations.” assumes an issue like “A support chatbot can submit payment refunds using the employee's broad API token without approval thresholds”, not the specific observed weakness “An internal coding assistant sends proprietary source files to a public model endpoint despite a restricted-data policy”; it leaves the evidence goal unmet: test uploads of restricted repositories are blocked and logged without transferring code externally.

D:

“Add adversarial evaluation for rate exhaustion, sensitive output, prompt injection, and model extraction” addresses “A team tests an AI model using only ordinary prompts, omitting model denial-of-service and extraction abuse cases”, with “A test suite records controlled failure rates, safeguards, and remediation owners across attack categories” as its acceptance test. Here the specific problem is “A cross-functional security group is considering a decision: An internal coding assistant sends proprietary source files to a public model endpoint despite a restricted-data policy”: it requires “Enforce approved model endpoints, source-code classification checks, and egress control on AI integrations” and evidence that “Test uploads of restricted repositories are blocked and logged without transferring code externally”.

Question 110

Which test or documented result best shows that the underlying issue was properly addressed? An internal coding assistant sends proprietary source files to a public model endpoint despite a restricted-data policy.

  1. Test uploads of restricted repositories are blocked and logged without transferring code externally.
  2. Training inputs exclude unnecessary identifiers and a privacy audit traces approved data sources.
  3. A release candidate has verifiable model hashes, supplier evidence, and traceable evaluation datasets.
  4. Access is blocked pending an authorized cross-border decision and supporting jurisdictional documentation.

Explanation

Correct Answer

A:

This is direct test evidence of the control objective in the case: An internal coding assistant sends proprietary source files to a public model endpoint despite a restricted-data policy. It demonstrates that the proposed response, enforce approved model endpoints, source-code classification checks, and egress control on AI integrations., produces the measurable result test uploads of restricted repositories are blocked and logged without transferring code externally. The observation tests a relevant security outcome rather than an unrelated control or reporting activity.

Incorrect Answers

B:

For the issue “An internal coding assistant sends proprietary source files to a public model endpoint despite a restricted-data policy”, measuring “Training inputs exclude unnecessary identifiers and a privacy audit traces approved data sources.” would validate “A model-training pipeline includes customer identifiers that are not necessary for the intended forecasting task” instead of the required control effect: test uploads of restricted repositories are blocked and logged without transferring code externally.

C:

This measurement is tied to “A healthcare AI model's external training supplier cannot identify where checkpoint weights and evaluation data were sourced” and the control “Require supplier provenance attestations, artifact integrity, and test-data chain of custody for model release”. It cannot confirm that “Enforce approved model endpoints, source-code classification checks, and egress control on AI integrations” has achieved “Test uploads of restricted repositories are blocked and logged without transferring code externally” for the issue in this question.

D:

Observation “Access is blocked pending an authorized cross-border decision and supporting jurisdictional documentation.” would be useful for “An international supplier proposes to move restricted research telemetry to a foreign support location without export review”, but it does not prove that the response to “An internal coding assistant sends proprietary source files to a public model endpoint despite a restricted-data policy” achieved the relevant outcome, test uploads of restricted repositories are blocked and logged without transferring code externally.