Question 350
A team wants Claude to create messages that hide material fees from vulnerable customers. What is appropriate?
- Proceed if the wording is grammatically correct.
- Target only customers least likely to complain.
- Reject the deceptive approach and propose clear, accurate communication.
- Hide the fees in increasingly obscure language.
Explanation
Correct Answer
C:Assistance should not be used to exploit an audience’s vulnerabilities or conceal information necessary for an informed decision.
Incorrect Answers
A:Language quality does not make deceptive omissions acceptable or resolve the ethical issue in the requested communication.
B:Selecting a vulnerable audience does not legitimize deception and can intensify the harm caused by the practice.
D:Obscurity worsens the manipulation by making material terms harder for customers to understand and compare.
Question 351
A user reports an error, but the source actually supports Claude’s answer. What should happen?
- Change the answer immediately to agree with every complaint.
- Change the supported answer solely because the complainant is more senior.
- Verify the complaint against evidence and clarify the misunderstanding before changing the workflow.
- Ignore all future feedback from that user.
Explanation
Correct Answer
C:Feedback is valuable but not automatically correct; changes should address established defects rather than replace a supported answer with an unsupported preference.
Incorrect Answers
A:Uncritical acceptance can introduce errors and make the workflow inconsistent with its authoritative sources or governing requirements.
B:Seniority does not change what the authoritative source establishes; the complaint still needs evidence-based verification.
D:One misunderstanding does not establish that later reports lack value or that the user’s broader needs are irrelevant.
Question 352
An associate uses Claude to evaluate a supplier in which they have a financial interest. What remains necessary?
- Disclose the conflict and follow the organization’s decision process.
- Remove the evaluator’s name from the document.
- The model’s involvement automatically makes the evaluation impartial.
- Ask Claude to write that no conflict exists.
Explanation
Correct Answer
A:AI assistance does not remove the human conflict or guarantee that criteria, evidence selection, and interpretation are independent.
Incorrect Answers
B:Anonymity does not resolve the conflict or provide the organization with information needed to manage it appropriately.
C:The user still influences inputs and decision framing, so model participation does not neutralize undisclosed personal interests.
D:Generated wording cannot change the actual relationship or substitute for a truthful disclosure to responsible reviewers.
Question 353
A user requests a verified worldwide legal review in ten minutes using no external sources. What should Claude be prompted to do?
- Ignore the time and source constraints silently.
- Explain the conflict and offer bounded scope.
- Invent citations to satisfy the verification requirement.
- Label a general summary a verified global review.
Explanation
Correct Answer
B:The requested verification cannot be supported under those conditions. A bounded alternative preserves usefulness without falsely claiming comprehensive legal assurance.
Incorrect Answers
A:Unannounced scope changes prevent the user from understanding what was actually delivered.
C:Fabricated references create apparent assurance without actual legal support.
D:The label would overstate both evidence and jurisdictional coverage.
Question 354
Claude’s recommendations are accurate but impossible for the team to implement within its authority. What needs refinement?
- Match actions to authority and resources.
- Remove all implementation context.
- Assign higher privileges to everyone without approval.
- Keep the actions because they are theoretically sound.
Explanation
Correct Answer
A:A recommendation must be feasible for its intended owners or explicitly identify who can authorize the necessary change.
Incorrect Answers
B:Less context makes feasibility harder to assess and leaves the mismatch between recommendation and authority unresolved.
C:Broadening access is a consequential governance decision, not an automatic way to rescue an impractical recommendation.
D:Technical plausibility does not establish that the named team has the authority or resources to execute them.
Question 355
A newer model becomes available after a workflow has passed evaluation. What is appropriate before switching?
- Switch automatically because newer always means compatible.
- Retest only a single uncomplicated request that both models handled previously.
- Retest the candidate on representative cases.
- Discard the acceptance criteria before comparing models.
Explanation
Correct Answer
C:Newer capability does not guarantee unchanged behavior, formatting, latency, or review burden for an established workflow.
Incorrect Answers
A:Version changes can alter behavior and do not establish that all existing requirements remain satisfied without testing.
B:A narrow easy case does not cover the established workflow’s complexity, known failures, output contract, or performance constraints.
D:Removing the standard prevents a meaningful assessment of whether the candidate improves or regresses required behavior.
Question 356
A consultant repeatedly prepares materials for one client using the same approved background and guidance. What feature fits?
- Research mode as a substitute for client instructions.
- A dedicated, maintained client Project.
- A public artifact containing all confidential background.
- One unrelated chat per task with no shared setup.
Explanation
Correct Answer
B:Persistent project context supports related work without requiring the consultant to reconstruct the same approved background for each task.
Incorrect Answers
A:Research can gather evidence but does not establish the client’s approved internal requirements or recurring working conventions.
C:A published deliverable is not an appropriate substitute for a controlled workspace holding sensitive client reference material.
D:That approach loses the reusable context and increases repeated preparation for the same ongoing client engagement.
Question 357
A scanned table produces nonsensical values in Claude’s extraction. What should be checked first?
- The scan quality and extracted text or OCR representation.
- Assume every strange value is an intentional business exception.
- Only the tone of the extraction prompt.
- Reformat the extracted numbers before checking them against the scan.
Explanation
Correct Answer
A:If the input representation is corrupted, the model may be reasoning over incorrect characters; the source needs correction or manual verification.
Incorrect Answers
B:Unexpected results should be checked against the original scan before being accepted as genuine operational data.
C:Tone does not restore unreadable digits or correct characters that were misrecognized before analysis.
D:Tidier display can make corrupted input appear credible while leaving the underlying character-recognition errors unresolved.
Question 358
Each of six documents fits individually within a workflow's documented input capacity, but all six together plus the conversation exceed it. What should the associate assess?
- Only the size of the largest individual document.
- Whether each document can be renamed to a shorter title.
- The combined context footprint and a feasible input allocation.
- Whether a promise of a short answer makes all input fit.
Explanation
Correct Answer
C:Capacity applies to the material processed together; separate acceptability of each document does not establish that the entire combined request fits.
Incorrect Answers
A:The combined request contains the other documents and conversation too, so the largest file alone understates its footprint.
B:Shorter titles do not remove the substantive document content responsible for exceeding the combined input capacity.
D:A shorter requested output does not automatically make an oversized input collection fit the workflow's documented limits.
Question 359
A difficult synthesis requires substantial reasoning, and additional latency is acceptable. Which family should be evaluated as the high-capability candidate, without assuming its answers are automatically correct?
- Opus.
- Haiku, on the assumption that the fastest family necessarily reasons most deeply.
- Sonnet, because balanced positioning excludes Opus from complex tasks.
- No family, because a model can never assist with complex synthesis.
Explanation
Correct Answer
A:Opus is commonly positioned for demanding reasoning. The team should still verify factual support and performance on the intended synthesis task.
Incorrect Answers
B:Efficiency and reasoning depth are separate dimensions; speed alone does not establish the demanding capability requested here.
C:Sonnet can be useful, but its balanced positioning does not exclude the high-capability family from consideration for demanding analysis.
D:Complexity creates a need for evaluation and review, not a blanket inability to assist with synthesizing supplied evidence.
Question 360
A Claude-assisted workflow is slow, but generation itself is fast. What should be measured?
- Only the first word’s appearance time.
- Only the model’s advertised speed.
- Time spent retrieving sources, waiting for reviews, formatting, and transferring outputs.
- Record only the total elapsed time without identifying the contributing stages.
Explanation
Correct Answer
C:End-to-end diagnosis should locate the real delay rather than assume model generation dominates every workflow’s elapsed time.
Incorrect Answers
A:An early response token does not establish when the complete accepted deliverable becomes available to its users.
B:General positioning does not reveal delays introduced by retrieval, coordination, or downstream steps in this actual process.
D:A total confirms the symptom but does not locate whether retrieval, review, formatting, or another stage accounts for the delay.