Anthropic CCAR-P (Claude Certified Architect - Professional) Practice Test 36

Question 350

A support log contains masked account information in normal use but exposes unmasked details on retry failures. What independently verifiable observation is needed?

  1. The verified result is that the adversarial suite records appropriate refusal without blocking benign informational requests.
  2. The recorded test establishes that routine support traces contain no prohibited identifiers under role-based access tests.
  3. The representative case test finds that injected pages fail to elicit secrets even when the model repeats their instructions.
  4. The staged rollout demonstrates that red-team runs fail at backend enforcement even if model text is manipulated.

Explanation

Correct Answer

B:

Input constraints and output handling both matter because sensitive data can leak through secondary systems; the test is necessary where a restricted assistant sometimes reveals private details in generated support logs.

Incorrect Answers

A:

A scoped safety policy should prevent harmful responses while preserving legitimate help; this makes sense when a public service receives harassment and self-harm requests outside its permitted purpose, rather than the risk that a support log contains masked account information in normal use but exposes unmasked details on retry failures.

C:

Layered controls prevent source text from overriding authority and block actual secret access; the trigger would be that a retrieved page tells an assistant to reveal a system secret during a normal summary, while the immediate question involves a support log contains masked account information in normal use but exposes unmasked details on retry failures.

D:

No single natural-language instruction can replace controls at retrieval and action boundaries; the design helps when a security review finds the team relies only on a prompt asking the model to obey policy, but the present constraint is that a support log contains masked account information in normal use but exposes unmasked details on retry failures.

Question 351

When the authorization service times out, a case-approval agent continues as if consent were granted. What outcome would demonstrate this requirement in operation?

  1. The readiness assessment shows that fault injection shows denied commits until the policy dependency is restored.
  2. The independent test establishes that routine support traces contain no prohibited identifiers under role-based access tests.
  3. The acceptance evidence shows that injected pages fail to elicit secrets even when the model repeats their instructions.
  4. The recorded test establishes that the adversarial suite records appropriate refusal without blocking benign informational requests.
  5. The comparison run confirms that malformed and extra-field payloads are blocked before reaching the executor.

Explanation

Correct Answer

A:

A missing control should not silently convert a high-stakes rejection into approval; an acceptance test needs to cover cases where a sensitive workflow loses its policy service and continues approving cases by default.

Incorrect Answers

B:

Input constraints and output handling both matter because sensitive data can leak through secondary systems; this applies when a restricted assistant sometimes reveals private details in generated support logs, not this situation where when the authorization service times out, a case-approval agent continues as if consent were granted.

C:

Layered controls prevent source text from overriding authority and block actual secret access; the alternative helps if a retrieved page tells an assistant to reveal a system secret during a normal summary, whereas this scenario involves when the authorization service times out, a case-approval agent continues as if consent were granted.

D:

A scoped safety policy should prevent harmful responses while preserving legitimate help; it is justified if a public service receives harassment and self-harm requests outside its permitted purpose, but here when the authorization service times out, a case-approval agent continues as if consent were granted.

E:

Structured syntax alone does not establish that content is safe for downstream execution; its scope includes cases where a model generates seemingly valid JSON containing an unexpected shell-command field, instead of when when the authorization service times out, a case-approval agent continues as if consent were granted.

Question 352

A tool's JSON parser accepts an unfamiliar field that will later be interpreted as executable shell text. Which observed test outcome would justify sign-off?

  1. The staged rollout demonstrates that routine support traces contain no prohibited identifiers under role-based access tests.
  2. The audit sample demonstrates that malformed and extra-field payloads are blocked before reaching the executor.
  3. The readiness assessment shows that red-team runs fail at backend enforcement even if model text is manipulated.
  4. The independent test establishes that fault injection shows denied commits until the policy dependency is restored.

Explanation

Correct Answer

B:

Structured syntax alone does not establish that content is safe for downstream execution; the observed protection matters when a model generates seemingly valid JSON containing an unexpected shell-command field.

Incorrect Answers

A:

Input constraints and output handling both matter because sensitive data can leak through secondary systems; it serves conditions where a restricted assistant sometimes reveals private details in generated support logs, but this decision concerns a tool's JSON parser accepts an unfamiliar field that will later be interpreted as executable shell text.

C:

No single natural-language instruction can replace controls at retrieval and action boundaries; this makes sense when a security review finds the team relies only on a prompt asking the model to obey policy, but the requested boundary concerns a tool's JSON parser accepts an unfamiliar field that will later be interpreted as executable shell text.

D:

A missing control should not silently convert a high-stakes rejection into approval; this technique fits where a sensitive workflow loses its policy service and continues approving cases by default, not the present case where a tool's JSON parser accepts an unfamiliar field that will later be interpreted as executable shell text.

Question 353

A safety demonstration relies on reassuring system-prompt wording instead of enforcing restrictions at tools. Which sign-off finding best checks the actual boundary?

  1. The pilot evaluation finds that red-team runs fail at backend enforcement even if model text is manipulated.
  2. The comparison run confirms that the adversarial suite records appropriate refusal without blocking benign informational requests.
  3. The staged rollout demonstrates that fault injection shows denied commits until the policy dependency is restored.
  4. The audit sample demonstrates that routine support traces contain no prohibited identifiers under role-based access tests.

Explanation

Correct Answer

A:

No single natural-language instruction can replace controls at retrieval and action boundaries; the test is necessary where a security review finds the team relies only on a prompt asking the model to obey policy.

Incorrect Answers

B:

A scoped safety policy should prevent harmful responses while preserving legitimate help; that is defensible if a public service receives harassment and self-harm requests outside its permitted purpose, while the immediate question involves a safety demonstration relies on reassuring system-prompt wording instead of enforcing restrictions at tools.

C:

A missing control should not silently convert a high-stakes rejection into approval; the other workload is one where a sensitive workflow loses its policy service and continues approving cases by default, but the present constraint is that a safety demonstration relies on reassuring system-prompt wording instead of enforcing restrictions at tools.

D:

Input constraints and output handling both matter because sensitive data can leak through secondary systems; this applies when a restricted assistant sometimes reveals private details in generated support logs, whereas the current issue is that a safety demonstration relies on reassuring system-prompt wording instead of enforcing restrictions at tools.

Question 354

A clinical note draft will be written into the official patient record. Which option supplies the needed architectural boundary?

  1. In the initial solution, use risk-tiered review and reserve mandatory approval for consequential exceptions.
  2. In the proposed design, capture reviewer edits and reasons in a governed feedback-to-evaluation process.
  3. Under the reviewed approach, place human review before issuing consequential adverse-action communications.
  4. For this production path, require qualified clinician approval before making the generated note authoritative.

Explanation

Correct Answer

D:

Clinical accountability needs a human review gate at the material publication boundary; validation needs to show that unapproved drafts cannot be committed to the medical record.

Incorrect Answers

A:

Proportionate oversight keeps throughput viable without removing controls on high-impact decisions; that belongs to situations where a low-risk internal FAQ reply is escalated for human approval every time, causing severe delays, not this situation where a clinical note draft will be written into the official patient record.

B:

Oversight should improve the system through traceable, reviewed examples rather than only patching one case; it is relevant if a human reviewer can approve or reject outputs but cannot correct recurring errors, whereas this scenario involves a clinical note draft will be written into the official patient record.

C:

High-stakes outcomes should not be finalized from unverified language-model inferences; it serves conditions where a loan assistant proposes denial notices that could affect legal rights, but here a clinical note draft will be written into the official patient record.

Question 355

A low-risk internal FAQ reply is escalated for human approval every time, causing severe delays. Which action addresses the root concern without unnecessary scope?

  1. At this decision point, present source evidence, uncertainty indicators and the specific proposed action for review.
  2. As a candidate control, require scoped, time-bound approvals linked to the exact side effect and amount.
  3. In the initial solution, route by impact, uncertainty and policy exceptions with explicit escalation thresholds.
  4. In the proposed design, use risk-tiered review and reserve mandatory approval for consequential exceptions.

Explanation

Correct Answer

D:

Proportionate oversight keeps throughput viable without removing controls on high-impact decisions; the practical acceptance condition is that routine answers pass automated checks while exceptional cases enter a reviewer queue.

Incorrect Answers

A:

Human oversight is ineffective when reviewers cannot inspect the basis of the decision; this control is needed when a reviewer sees a generated recommendation but no evidence or reason for escalation, but the actual risk is that a low-risk internal FAQ reply is escalated for human approval every time, causing severe delays.

B:

Broad standing consent fails to constrain later actions whose risk and parameters differ; that would be useful if an agent asks for approval once at login and then performs unlimited future purchases, but the requested boundary concerns a low-risk internal FAQ reply is escalated for human approval every time, causing severe delays.

C:

Triage preserves expert capacity for the cases where judgment changes risk; it serves a setting where a moderation queue grows without bound despite only some cases having meaningful ambiguity, not the present case where a low-risk internal FAQ reply is escalated for human approval every time, causing severe delays.

Question 356

A loan assistant proposes denial notices that could affect legal rights. What is the most appropriate initial architecture decision?

  1. During architecture review, place human review before issuing consequential adverse-action communications.
  2. In the proposed implementation, present source evidence, uncertainty indicators and the specific proposed action for review.
  3. At this decision point, require scoped, time-bound approvals linked to the exact side effect and amount.
  4. As a candidate control, use risk-tiered review and reserve mandatory approval for consequential exceptions.

Explanation

Correct Answer

A:

High-stakes outcomes should not be finalized from unverified language-model inferences; the controlled test needs to find that a denied application generates no customer notice until an authorized reviewer signs off.

Incorrect Answers

B:

Human oversight is ineffective when reviewers cannot inspect the basis of the decision; the approach helps when a reviewer sees a generated recommendation but no evidence or reason for escalation, but the present constraint is that a loan assistant proposes denial notices that could affect legal rights.

C:

Broad standing consent fails to constrain later actions whose risk and parameters differ; that belongs to situations where an agent asks for approval once at login and then performs unlimited future purchases, whereas the current issue is that a loan assistant proposes denial notices that could affect legal rights.

D:

Proportionate oversight keeps throughput viable without removing controls on high-impact decisions; it is relevant if a low-risk internal FAQ reply is escalated for human approval every time, causing severe delays, whereas the release criterion addresses a loan assistant proposes denial notices that could affect legal rights.

Question 357

A reviewer sees a generated recommendation but no evidence or reason for escalation. Which architecture change should be approved first?

  1. As a scoped first step, use risk-tiered review and reserve mandatory approval for consequential exceptions.
  2. In the release design, require scoped, time-bound approvals linked to the exact side effect and amount.
  3. During architecture review, capture reviewer edits and reasons in a governed feedback-to-evaluation process.
  4. In the proposed implementation, require qualified clinician approval before making the generated note authoritative.
  5. In the implementation sequence, present source evidence, uncertainty indicators and the specific proposed action for review.

Explanation

Correct Answer

E:

Human oversight is ineffective when reviewers cannot inspect the basis of the decision; the intended outcome is that reviewer studies show unsupported conclusions are identified before approval.

Incorrect Answers

A:

Proportionate oversight keeps throughput viable without removing controls on high-impact decisions; the design helps when a low-risk internal FAQ reply is escalated for human approval every time, causing severe delays, whereas this scenario involves a reviewer sees a generated recommendation but no evidence or reason for escalation.

B:

Broad standing consent fails to constrain later actions whose risk and parameters differ; that fits cases where an agent asks for approval once at login and then performs unlimited future purchases, but here a reviewer sees a generated recommendation but no evidence or reason for escalation.

C:

Oversight should improve the system through traceable, reviewed examples rather than only patching one case; this control is needed when a human reviewer can approve or reject outputs but cannot correct recurring errors, instead of when a reviewer sees a generated recommendation but no evidence or reason for escalation.

D:

Clinical accountability needs a human review gate at the material publication boundary; that would be useful if a clinical note draft will be written into the official patient record, not the case in which a reviewer sees a generated recommendation but no evidence or reason for escalation.

Question 358

A moderation queue grows without bound despite only some cases having meaningful ambiguity. What should be incorporated into the approved design?

  1. For the operating team, require qualified clinician approval before making the generated note authoritative.
  2. For this deployment, route by impact, uncertainty and policy exceptions with explicit escalation thresholds.
  3. As a scoped first step, capture reviewer edits and reasons in a governed feedback-to-evaluation process.
  4. For the current integration, require scoped, time-bound approvals linked to the exact side effect and amount.

Explanation

Correct Answer

B:

Triage preserves expert capacity for the cases where judgment changes risk; production testing should demonstrate that queue metrics show high-risk coverage without mandatory review of every harmless item.

Incorrect Answers

A:

Clinical accountability needs a human review gate at the material publication boundary; it is justified if a clinical note draft will be written into the official patient record, but the requested boundary concerns a moderation queue grows without bound despite only some cases having meaningful ambiguity.

C:

Oversight should improve the system through traceable, reviewed examples rather than only patching one case; it targets cases where a human reviewer can approve or reject outputs but cannot correct recurring errors, while this problem involves a moderation queue grows without bound despite only some cases having meaningful ambiguity.

D:

Broad standing consent fails to constrain later actions whose risk and parameters differ; the approach helps when an agent asks for approval once at login and then performs unlimited future purchases, and not where a moderation queue grows without bound despite only some cases having meaningful ambiguity.

Question 359

An agent asks for approval once at login and then performs unlimited future purchases. Which approach should the design team choose?

  1. As the design priority, place human review before issuing consequential adverse-action communications.
  2. In the implementation sequence, require scoped, time-bound approvals linked to the exact side effect and amount.
  3. For the operating team, route by impact, uncertainty and policy exceptions with explicit escalation thresholds.
  4. For this deployment, use risk-tiered review and reserve mandatory approval for consequential exceptions.

Explanation

Correct Answer

B:

Broad standing consent fails to constrain later actions whose risk and parameters differ; the go-live check should establish that new or enlarged purchases require fresh authorization before execution.

Incorrect Answers

A:

High-stakes outcomes should not be finalized from unverified language-model inferences; this technique fits where a loan assistant proposes denial notices that could affect legal rights, but the present constraint is that an agent asks for approval once at login and then performs unlimited future purchases.

C:

Triage preserves expert capacity for the cases where judgment changes risk; the design helps when a moderation queue grows without bound despite only some cases having meaningful ambiguity, whereas the release criterion addresses an agent asks for approval once at login and then performs unlimited future purchases.

D:

Proportionate oversight keeps throughput viable without removing controls on high-impact decisions; that fits cases where a low-risk internal FAQ reply is escalated for human approval every time, causing severe delays, while the required safeguard concerns an agent asks for approval once at login and then performs unlimited future purchases.

Question 360

A human reviewer can approve or reject outputs but cannot correct recurring errors. Where should the solution architect focus the implementation?

  1. For the current integration, present source evidence, uncertainty indicators and the specific proposed action for review.
  2. Under the system plan, place human review before issuing consequential adverse-action communications.
  3. As the design priority, capture reviewer edits and reasons in a governed feedback-to-evaluation process.
  4. In the implementation sequence, require qualified clinician approval before making the generated note authoritative.

Explanation

Correct Answer

C:

Oversight should improve the system through traceable, reviewed examples rather than only patching one case; one operational check should confirm that recurring corrections become approved regression cases in the next release cycle.

Incorrect Answers

A:

Human oversight is ineffective when reviewers cannot inspect the basis of the decision; this applies when a reviewer sees a generated recommendation but no evidence or reason for escalation, but here a human reviewer can approve or reject outputs but cannot correct recurring errors.

B:

High-stakes outcomes should not be finalized from unverified language-model inferences; the alternative helps if a loan assistant proposes denial notices that could affect legal rights, instead of when a human reviewer can approve or reject outputs but cannot correct recurring errors.

D:

Clinical accountability needs a human review gate at the material publication boundary; its scope includes cases where a clinical note draft will be written into the official patient record, whereas the problem presented is that a human reviewer can approve or reject outputs but cannot correct recurring errors.

Claude Certified Architect - Professional (CCAR-P) Practice Tests