Anthropic CCAR-P (Claude Certified Architect - Professional) Practice Test 29

Question 280

A document-fetch tool accepts an approved public URL, which redirects to a private infrastructure address. Initial-host checks pass. Which change closes this specific gap?

  1. Check only whether the original URL begins with HTTPS
  2. Validate the resolved destination and every redirect hop before fetching
  3. Increase the redirect limit so the request can finish
  4. Ask the model to describe the destination before invoking the tool

Explanation

Correct Answer

B:

A permitted starting URL can lead to a prohibited destination. Validation must cover the actual network targets, including redirects and address resolution.

Incorrect Answers

A:

Transport encryption does not determine whether a destination is authorized. A secure URL can still redirect to private infrastructure.

C:

A larger limit expands the chain the fetcher will follow. It does not protect the boundary between public content and private addresses.

D:

A generated description cannot reliably establish runtime DNS or redirect behavior. The fetch implementation must enforce the destination policy itself.

Question 281

An agent must run customer-supplied Python for a grading service. The code may be hostile. Which environment is most appropriate?

  1. A shared developer shell with a prompt asking for safe execution
  2. A production worker that has administrator rights for compatibility
  3. A separate directory inside the same unrestricted process
  4. An isolated disposable runtime with restricted resources, mounts, and network access

Explanation

Correct Answer

D:

Customer code should execute within enforced boundaries that limit its effects. Disposal and resource controls reduce persistence and denial-of-service exposure.

Incorrect Answers

A:

The prompt cannot constrain arbitrary customer code once it runs. A shared shell exposes the developer's accessible files, credentials, and processes.

B:

Broad privileges increase the consequences of hostile code. Compatibility should be provided within a constrained runtime rather than by granting unrestricted access.

C:

Directory organization alone does not isolate operating-system capabilities. The process could still access other paths, networks, and shared resources.

Question 282

An internally approved automation skill is updated by an upstream publisher. The new version adds a command that exports files. What should deployment governance require?

  1. Apply the update because the earlier version was approved
  2. Skip review when the publisher labels the release a patch
  3. Allow the update if its documentation uses the same title
  4. Review the changed version and its capabilities before replacing the approved release

Explanation

Correct Answer

D:

Approval applies to the reviewed behavior and permissions, not indefinitely to a package name. A new export capability changes the risk assessment.

Incorrect Answers

A:

Prior approval does not cover new capabilities automatically. The update introduces behavior that the original review did not assess.

B:

Release labels do not prove that security-relevant behavior is unchanged. The added export command still requires review under the organization's process.

C:

A title is not an integrity or behavior guarantee. The actual version and changes determine whether the existing approval remains applicable.

Question 283

Reviewers see an identical confirmation dialog for opening a harmless file and sending a confidential report externally. They approve almost every dialog without reading. What redesign best addresses the problem?

  1. Show the model's confidence score instead of the proposed action
  2. Remove confirmations because reviewers rarely reject them
  3. Require a generic acknowledgment at the start of every session
  4. Reserve prominent approval for consequential actions and show their specific effects
  5. Require two identical confirmation dialogs for every action

Explanation

Correct Answer

D:

Risk-sensitive prompts preserve reviewer attention for meaningful decisions. Showing the recipient, data, and consequences supports informed approval of the external send.

Incorrect Answers

A:

Confidence does not specify what information will leave or who will receive it. Reviewers need the concrete action and its risk-relevant details.

B:

Low rejection rates do not prove the actions are safe. The observed behavior indicates a review design problem that should be corrected.

C:

A session-level acknowledgment does not expose the consequences of a later sensitive transfer. The review needs to be tied to the consequential action.

E:

Duplicating low-information prompts adds friction without improving the evidence. It may reinforce the same automatic approval behavior.

Question 284

A human reviewer must verify a generated contract summary but sees only the summary and an Approve button. Which interface addition most directly enables substantive review?

  1. Linked source passages and visible uncertainty for the claims under review
  2. A timer encouraging approval within a few seconds
  3. A badge indicating that the model is highly capable
  4. A summary of previous reviewers' approval rates

Explanation

Correct Answer

A:

Reviewers need accessible evidence to evaluate the summary's claims. Source links and uncertainty help them identify unsupported or ambiguous interpretations.

Incorrect Answers

B:

A speed target does not make the review more informed. It can reduce the time available to inspect complex or ambiguous contract details.

C:

A capability label does not establish correctness for this contract. The reviewer still lacks the evidence needed to check individual claims.

D:

Past approval frequency does not show whether this contract summary is supported. The reviewer needs evidence for the claims currently being assessed.

Question 285

An organization permits an assistant to recommend account restrictions, with an employee making the final decision. Affected users cannot dispute incorrect source records. What governance gap remains?

  1. A longer retention period for decisions without a correction process
  2. A second employee approval that uses the same unchallengeable source records
  3. A higher model-confidence threshold for recommending restrictions
  4. A documented path to challenge the evidence and correct the resulting decision

Explanation

Correct Answer

D:

Human approval alone does not provide recourse for affected users. A challenge process must reach the source error and any decision based on it.

Incorrect Answers

A:

Keeping a record does not provide a way to challenge inaccurate evidence. Retention and redress serve different purposes.

B:

Another approval may repeat the same error if the evidence cannot be corrected. The gap concerns affected users' access to a challenge and correction path.

C:

Confidence does not establish that an underlying record is accurate. It cannot replace a route for disputing and correcting the evidence.

Question 286

A customer service assistant appears to be a human representative and cannot reliably recognize when a customer needs escalation. What pair of changes most directly supports informed use? Choose TWO.

  1. Provide an accessible route to a qualified human when needed
  2. Clearly identify the automated assistant and its relevant limitations
  3. Describe every internal infrastructure component in the greeting
  4. Use a human photograph and personal biography for the assistant
  5. Treat continued conversation as approval for any later action

Explanation

Correct Answers

A:

A usable escalation path gives customers recourse when automation cannot meet their needs. Disclosure alone does not supply that alternative.

B:

Users need an accurate understanding of the service they are interacting with. A clear disclosure supports informed expectations about the assistant's role.

Incorrect Answers

C:

Implementation details are not a substitute for understandable disclosure. A lengthy component list may obscure the assistant's role and limitations.

D:

A fabricated human identity would reinforce the misunderstanding. It does not support informed expectations about an automated interaction.

E:

Continuing a chat does not specify consent to every future consequence. The system still needs appropriate authorization for consequential actions.

Question 287

A governance team reviews registered assistants but discovers unregistered pilots processing company documents in several departments. What capability is missing?

  1. A list of approved model names without recording deployed use cases
  2. A rule that only production systems can appear in the register
  3. More frequent reviews of only the already registered assistants
  4. An inventory and intake process that captures AI uses before relevant review gates

Explanation

Correct Answer

D:

Controls cannot be applied consistently to systems the organization does not know exist. Intake and ownership make pilots visible at the stage when their scope and data use are assessed.

Incorrect Answers

A:

Model names do not identify the departments, data, purposes, or authority of each deployment. Governance needs an inventory of uses, not merely vendors.

B:

Pilots can already process consequential data or perform actions. Excluding them preserves the exact visibility gap identified in the scenario.

C:

Frequency improves oversight of known systems but does not discover the omitted pilots. The inventory boundary remains incomplete.

Question 288

A routing model never receives a protected attribute, yet a neighborhood code strongly influences which users receive manual assistance. What is the best next investigation?

  1. Test for proxy effects and relevant group outcome differences
  2. Compare only the overall routing accuracy across all users
  3. Replace neighborhood codes with opaque identifiers and keep the same mapping
  4. Conclude that disparate effects are impossible without the attribute
  5. Ask the model whether it considers itself unbiased

Explanation

Correct Answer

A:

Excluding an explicit attribute does not eliminate correlated signals. The investigation should connect feature use to observed outcomes and the service's fairness objectives.

Incorrect Answers

B:

Aggregate accuracy can conceal uneven access to assistance. The relevant investigation needs appropriate subgroup and feature-level evidence.

C:

Renaming a signal does not remove the information it carries. The same mapping can preserve the same disparity under less interpretable labels.

D:

Other variables can correlate with protected characteristics. Absence of the explicit attribute is insufficient evidence that the outcomes are fair.

E:

A self-description does not measure actual decisions or their effects. Fairness assessment requires evidence from the system's behavior and data.

Question 289

A voice-only assistant replaces a text support channel. Some customers cannot use speech input. What deployment change best addresses this exclusion?

  1. Increase the speech recognition model's average benchmark score
  2. A text form limited to requesting a callback by phone
  3. Longer voice-response time limits for all callers
  4. Provide an accessible alternative channel with comparable service access

Explanation

Correct Answer

D:

The barrier is the interaction modality, so users need a usable way to obtain the service. An alternative should preserve meaningful access rather than merely document the exclusion.

Incorrect Answers

A:

Better recognition may help spoken interactions but does not enable users who cannot provide speech. The modality barrier remains.

B:

A text entry point that leads only to voice support does not provide comparable access for customers who cannot use speech. The alternative must support the actual service.

C:

More time can help some callers but does not remove a speech-only entry requirement. Customers unable to provide speech still need an accessible alternative.

Question 290

A company collects chats to resolve support cases. A team proposes reusing them to infer employee performance without an approved purpose or data review. What should happen first?

  1. Assess the proposed new purpose, permissions, and applicable governance requirements
  2. Train the model first and seek approval only if results are useful
  3. Proceed because internal ownership permits any use of collected data
  4. Remove customer names and assume every secondary use is permitted

Explanation

Correct Answer

A:

A new use can change expectations, risks, and obligations even when the data already exists. The organization must establish whether that secondary use is authorized.

Incorrect Answers

B:

Model development would already use the data for the unapproved purpose. Review should precede that processing rather than depend on its apparent usefulness.

C:

Possession alone does not establish an unrestricted purpose. The proposed employee analysis requires its own justification and governance assessment.

D:

Removing names may reduce some risk but does not authorize a new purpose. Remaining content may also retain sensitive or identifying information.

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