Anthropic CCAR-P (Claude Certified Architect - Professional) Practice Test 31

Question 300

An autonomous workflow begins issuing harmful external updates. Its normal approval service is also unavailable. Which previously implemented control would provide the most dependable containment?

  1. An approval request sent through the unavailable approval service
  2. An independently operable switch that disables the workflow's write capability
  3. A dashboard that displays the number of harmful updates
  4. An instruction in the workflow asking it to stop after the current task

Explanation

Correct Answer

B:

Containment must remain available when an ordinary dependency fails. Disabling writes outside the affected workflow prevents further actions while responders investigate.

Incorrect Answers

A:

That path depends on the component already known to be unavailable. It cannot provide dependable immediate containment in this scenario.

C:

Visibility supports response but does not stop the actions. The team needs an enforceable control that can disable further writes.

D:

The affected workflow may not reliably honor that instruction. It also allows the current harmful task to continue before containment.

Question 301

Responders confirm that an assistant is exposing confidential records to unauthorized users, but the root cause is not yet known. What is the best immediate operational sequence?

  1. Erase all traces immediately and then restart every service
  2. Keep the access path open until the team proves the exact root cause
  3. Issue a general notice and defer technical action until the next release
  4. Contain the affected access path, preserve evidence, and investigate under the incident process

Explanation

Correct Answer

D:

Confirmed disclosure warrants limiting further exposure before completing diagnosis. Evidence preservation supports investigation without leaving the vulnerable path operating unrestricted.

Incorrect Answers

A:

Indiscriminate erasure can destroy evidence needed to understand scope and cause. Restarting alone also does not demonstrate that the disclosure path is closed.

B:

A complete diagnosis may take time while more records are disclosed. Containment can proceed based on the confirmed impact and known affected path.

C:

Communication does not contain ongoing unauthorized access. The active exposure requires an operational response before the normal release cycle.

Question 302

During an investigation, engineers want to copy raw confidential traces into an unrestricted shared document for convenience. What is the better evidence-handling approach?

  1. Preserve necessary evidence in a restricted incident workspace with controlled access and retention
  2. Copy the traces broadly so more people can volunteer to review them
  3. Replace every trace with the model's summary of what probably happened
  4. Delete all raw evidence and rely exclusively on staff recollection
  5. Keep unrestricted copies permanently in case they become useful

Explanation

Correct Answer

A:

Investigation needs evidence without creating another uncontrolled disclosure. A restricted workspace and defined lifecycle support both analysis and appropriate data handling.

Incorrect Answers

B:

Broad distribution increases exposure beyond the response team. Access should follow the investigation's actual need rather than general availability.

C:

A generated reconstruction can omit or invent decisive details. It should not replace preserved evidence used to establish incident facts.

D:

Recollection may omit details needed to determine cause and scope. Necessary evidence should be protected rather than discarded without an investigation plan.

E:

Indefinite broad retention creates additional risk without a defined purpose. Evidence handling should specify access and retention for the investigation.

Question 303

A prelaunch review lists several serious residual risks, but each has the owner field set to 'AI team.' What change most directly makes the register actionable?

  1. Remove residual risks once their initial mitigations are implemented
  2. Let the model assign severity scores without human review
  3. Assign accountable owners, treatment decisions, deadlines, and escalation criteria
  4. Add more detailed descriptions while leaving ownership unchanged

Explanation

Correct Answer

C:

A risk list needs explicit responsibility and follow-through to guide action. Named ownership and treatment criteria make unresolved risks visible to the appropriate decision makers.

Incorrect Answers

A:

Mitigation does not necessarily eliminate a risk. Residual exposure should remain visible until it is evaluated and appropriately accepted or treated.

B:

Scores alone do not create accountability or authorized treatment decisions. The organization still needs responsible owners and a defined risk process.

D:

Description can improve understanding but does not establish who must act. The identified gap is accountability and tracking of risk treatment.

Question 304

Security documents a residual data exposure risk that exceeds the project's normal threshold. A developer wants to mark it accepted to meet a release date. Who should decide?

  1. The designated authority with responsibility to accept that class of residual risk
  2. Any engineer who can edit the risk register
  3. The model, using a prompt containing the release deadline
  4. The vendor's sales representative for the model service

Explanation

Correct Answer

A:

Risk acceptance is an accountable organizational decision, not merely a technical status change. The designated authority must consider the documented exposure and business context.

Incorrect Answers

B:

Editing access does not establish authority to accept organizational risk. The decision should follow the assigned responsibility for this class of exposure.

C:

A generated recommendation cannot assume the organization's acceptance authority. The deadline is context, not a delegation of accountability.

D:

A vendor representative does not own the customer's application risk decision. Acceptance belongs to the customer's designated authority.

Question 305

An incident is contained by fixing a vulnerable prompt, but the agent's persistent memory still contains attacker-created instructions. What additional recovery step is necessary?

  1. Identify and remove or restore contaminated state before re-enabling its use
  2. Keep all memory because it predates the repaired deployment
  3. Increase the model's context window so it can understand the memory better
  4. Restore the newest memory snapshot without checking when contamination began

Explanation

Correct Answer

A:

The prompt fix prevents one entry path but does not clean existing malicious state. Recovery must address persistence that could reintroduce the behavior.

Incorrect Answers

B:

Earlier storage does not make attacker-created instructions trustworthy. The state is part of the incident's continuing effect and needs review.

C:

More context capacity does not establish the integrity of stored instructions. It may simply expose the agent to more contaminated material.

D:

The latest snapshot can already contain attacker-created instructions. Recovery needs a known suitable state or a targeted review of the affected content.

Question 306

An incident may affect several customers. Engineers have preliminary logs but incomplete scope. How should external notification decisions be handled?

  1. Use the established incident communication process with accountable legal, privacy, and business owners
  2. Wait for perfect certainty before informing any accountable internal owner
  3. Apply the notification deadline from the last incident without checking current obligations
  4. Have each engineer independently contact customers they recognize in logs

Explanation

Correct Answer

A:

Notification requires coordinated facts, responsibilities, and applicable obligations. The process should manage uncertainty explicitly while the technical investigation continues.

Incorrect Answers

B:

Internal escalation supports timely decisions while uncertainty remains. Perfect certainty may be unavailable and is not a prerequisite for activating the process.

C:

A past incident can differ in affected data, parties, jurisdictions, and contracts. Accountable owners must assess the present facts rather than reuse a deadline automatically.

D:

Independent outreach risks inconsistent or inaccurate statements and unnecessary disclosure. Communication should be coordinated through the authorized incident process.

Question 307

A support tool inserts model-generated text into a page using raw HTML. A retrieved document causes the output to include executable markup. What control addresses the rendering vulnerability?

  1. Increase the minimum answer length before displaying it
  2. Filter suspicious source documents while continuing to render arbitrary output as HTML
  3. Check only whether the answer contains a citation
  4. Render untrusted content safely using appropriate encoding or a constrained sanitization policy

Explanation

Correct Answer

D:

The browser must not interpret arbitrary generated text as executable markup. Safe rendering enforces that boundary independently of whether the model reproduces hostile content.

Incorrect Answers

A:

Longer output can contain the same executable content. Length is unrelated to whether the browser interprets the text as markup.

B:

Input filtering can reduce some attacks but does not enforce the browser boundary. Any remaining untrusted generated markup must still be rendered safely.

C:

A citation does not establish that the surrounding markup is safe. The rendering boundary still requires encoding or constrained sanitization.

Question 308

An assistant exports customer-provided descriptions to CSV. Some descriptions begin with spreadsheet formula characters, and users open the file in a spreadsheet application. What should the export design address?

  1. Assume CSV is safe because it contains no embedded macros
  2. Prevent formula interpretation of untrusted values during supported imports
  3. Quote all CSV fields and assume quoting disables spreadsheet formulas
  4. Ask the model to avoid discussing spreadsheet functions

Explanation

Correct Answer

B:

A CSV value can acquire executable spreadsheet meaning when opened. The exporter must apply a tested policy that preserves data without enabling unintended formula evaluation.

Incorrect Answers

A:

Formula interpretation does not require a macro-enabled file. The risk arises from how the spreadsheet application interprets imported cell values.

C:

CSV quoting primarily handles field parsing and separators. It does not necessarily stop the spreadsheet from interpreting a cell value as a formula.

D:

The dangerous value may originate directly from customer data. A topic instruction does not enforce safe serialization for the exported cells.

Question 309

A risk board compares a rare but severe data disclosure with frequent minor formatting defects. Which two dimensions must remain visible rather than being collapsed into incident count alone? Choose TWO.

  1. The raw number of historical incidents without considering exposure
  2. The likelihood or expected frequency under the deployment conditions
  3. The implementation cost of a mitigation treated as the failure's severity
  4. The ease of implementing a mitigation treated as the likelihood of failure
  5. The impact if each failure occurs

Explanation

Correct Answers

B:

Severity alone does not describe how often the organization may experience a failure. Likelihood should be estimated for the actual exposure and operating conditions.

E:

Frequency alone would understate rare failures with serious consequences. Impact describes the magnitude of harm relevant to treatment priorities.

Incorrect Answers

A:

Counts can differ because usage volume and observation periods differ. Likelihood assessment needs relevant exposure, while impact must remain a separate consideration.

C:

Mitigation cost informs treatment choices but does not describe the harm if the failure occurs. Severity and intervention expense should not be conflated.

D:

Implementation effort describes the treatment option, not how likely the underlying failure is. Likelihood should be estimated from deployment conditions and relevant evidence.

Question 310

A write tool must execute only after a positive authorization decision. Its authorization service times out, and no valid decision is available. Which behavior satisfies the stated contract?

  1. Reuse an unrelated user's last successful decision for the same tool
  2. Treat the model's confidence that access is legitimate as a substitute
  3. Execute once because the timeout is probably a temporary infrastructure failure
  4. Do not execute the write; return a distinct authorization-unavailable outcome

Explanation

Correct Answer

D:

A timeout supplies no positive permission. Separating unavailable authorization from an explicit denial supports recovery while preserving the required execution boundary.

Incorrect Answers

A:

Authorization is scoped to the relevant principal and request. A different user's decision cannot supply the missing permission for this write.

B:

A confidence statement is not the required authorization decision. The execution layer must honor its explicit dependency on the authorization service.

C:

A likely transient cause does not establish permission. The write would occur without the positive decision required by the contract.

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