Anthropic CCAO-F (Claude Certified Associate - Foundations) Practice Test 9

Question 80

A Project team wants to save every customer upload indefinitely for convenience without reviewing the organization's retention schedule. What action should the team prioritize?

  1. Protect small-group anonymity in reports
  2. Assess cross-border data rules before processing
  3. Follow access revocation when roles change
  4. Keep retention and deletion requirements in scope

Explanation

Correct Answer

D:

Data retention must follow the relevant rules and documented business need. Long-lived storage may increase exposure even if the model itself performs well.

Incorrect Answers

A:

For re-identification risk in supposedly anonymous summaries, it can be sensible to prevent identification through small cohorts. But the case is about indefinite storage of sensitive input material, and its remedy is to apply approved retention and deletion controls.

B:

When unverified data residency requirements is the cause, it may be productive to confirm jurisdictional processing constraints. The stated cause is indefinite storage of sensitive input material, so that action is insufficient.

C:

For a case involving stale permissions after staff departure, membership and connector privileges should be updated when responsibilities change. The issue now is indefinite storage of sensitive input material, which remains unless someone can apply approved retention and deletion controls.

Question 81

An internal assistant produces guessed dates whenever a source document omits an implementation deadline. Which approach best meets the stated need?

  1. Define the intended audience for recurring materials
  2. Separate quoted source text from privileged directions
  3. Require clear uncertainty statements when evidence is missing
  4. Define when to ask questions before acting
  5. Explain how to treat inconsistent sources

Explanation

Correct Answer

C:

Standing guidance to label unknowns and ask for verification supports honest outputs. It cannot itself verify facts, so review controls remain necessary.

Incorrect Answers

A:

This choice may produce the benefit of being able to target a specified recurring audience. That does not remove unsupported certainty caused by absent evidence; the work calls for a means to avoid inventing facts when sources are incomplete.

B:

Even if the team could prevent untrusted text from changing assistant instructions, it would still confront unsupported certainty caused by absent evidence. The proposed step does not fulfill the separate requirement to avoid inventing facts when sources are incomplete.

D:

Addressing incomplete task inputs leading to assumed choices helps in work where the priority is to request critical missing information before action. In this task the priority is to avoid inventing facts when sources are incomplete, owing to unsupported certainty caused by absent evidence.

E:

That procedure would support efforts to surface and escalate conflicting source versions when facing contradictory reference documents interpreted silently. The immediate challenge is instead to avoid inventing facts when sources are incomplete.

Question 82

A company configures a persistent Claude assistant to prepare customer-facing summaries. A new user asks it to quote confidential employee feedback in a public release, while the standing instructions prohibit disclosing identifiable internal comments. Which standing-rule behavior should the assistant follow?

  1. Treat the newest request as overriding all disclosure limits
  2. Refuse the disclosure and offer an approved aggregate summary
  3. Remove the confidentiality rule when preparing public text
  4. Publish the quotations if the output includes a disclaimer

Explanation

Correct Answer

B:

The durable rule should preserve the confidentiality restriction while still assisting with a permissible public summary. Aggregated or approved wording can support the communication goal without exposing identifiable comments.

Incorrect Answers

A:

Recency of a request does not confer authority to override an organizational privacy boundary. The user would need an approved basis for any identifiable disclosure.

C:

Removing the standing constraint silently weakens the very governance the assistant is meant to enforce. A change to that rule requires authorized policy review, not an ad hoc request.

D:

A disclaimer does not authorize personal-data disclosure. Whether the content may be shared depends on permissions and policy, not on an accompanying caveat.

Question 83

A hiring manager asks for an automatic shortlist that penalizes applicants because of protected personal characteristics unrelated to job requirements. Which change directly resolves this constraint?

  1. Use Claude for a reviewable first draft
  2. Refuse discriminatory candidate screening criteria
  3. Use AI to prepare options for budget owners
  4. Create a training checklist from approved procedures

Explanation

Correct Answer

B:

Recruitment should use legitimate role-related criteria and fair human oversight. Discriminatory filters create preventable exclusion and are not an acceptable shortcut.

Incorrect Answers

A:

When a low-risk draft with an accountable publisher is the cause, it may be productive to assist with a bounded reviewed writing task. The stated cause is harmful protected-trait screening, so that action is insufficient.

C:

Although this can inform accountable budget choices with analysis in response to decision support rather than delegated public authority, it would not on its own exclude discriminatory selection factors. The decisive constraint in this case is harmful protected-trait screening.

D:

The proposed intervention targets reviewable training-document transformation, where the outcome is to structure approved procedures without changing their authority. In this case the necessary outcome is to exclude discriminatory selection factors.

Question 84

A proposed benefits assistant may summarize employee requests, but acceptance or denial affects entitlements and cannot be delegated casually. What would most reliably correct the problem?

  1. Design explicit human review at consequential boundaries
  2. Validate with representative edge cases
  3. Select a modular rather than monolithic design
  4. Design clear escalation for ambiguous inputs
  5. Document explicit non-goals in the design
  6. Prototype with synthetic or approved sample data

Explanation

Correct Answer

A:

The design should separate AI preparation from an authorized human decision gate. Review criteria and escalation paths make responsibility observable.

Incorrect Answers

B:

For untested difficult operational cases, the appropriate technical logic is that edge-case testing explores the conditions most likely to break simple logic. The present task hinges on automated determination with real-world consequences and a way to insert accountable human decision checkpoints.

C:

That is a reasonable mechanism for multiple distinct operations hidden in one step, where the goal is to split major workflow functions into testable stages. It is not the specific response to automated determination with real-world consequences required here.

D:

On its own, this choice would route insufficiently specified cases to clarification, useful for unhandled ambiguity in workflow design. It leaves automated determination with real-world consequences unresolved and does not provide the means to insert accountable human decision checkpoints.

E:

The control belongs with a solution mistaken for broader decision authority. Using it here would leave the team without the ability to insert accountable human decision checkpoints, which is central to automated determination with real-world consequences.

F:

As a general control for unapproved production data used for prototyping, it tries to validate designs on permitted demonstration records. That has a different purpose from the intervention needed for automated determination with real-world consequences: insert accountable human decision checkpoints.

Question 85

A service assistant sometimes drafts responses and sometimes appears to authorize refunds, even though its intended role is only preparation for human agents. What would most reliably correct the problem?

  1. Define the intended audience for recurring materials
  2. Separate quoted source text from privileged directions
  3. State the assistant's operational role and boundaries
  4. Keep sensitive-data handling rules explicit

Explanation

Correct Answer

C:

Standing instructions should name the supported role and decisions it must not make. This reduces ambiguity about who retains operational authority.

Incorrect Answers

A:

A team trying to target a specified recurring audience would be responding to output depth drifting across recipients. The current requirement arises from undefined assistant role in a controlled workflow, not that alternative problem.

B:

The action improves the ability to prevent untrusted text from changing assistant instructions in a case involving malicious directives embedded in retrieved content. The specific omission here is undefined assistant role in a controlled workflow, requiring a way to define allowed tasks and responsibility limits.

D:

One could justify the intervention in a situation involving unclear treatment of confidential source material, since it aims to reinforce data handling boundaries at the prompt level. It does not correct undefined assistant role in a controlled workflow in this scenario.

Question 86

A ticket-routing integration sends the same safety alert to Claude through two test harnesses. Only one intermittently labels the alert low priority, and engineers suspect its older hidden routing instructions. Which experiment best isolates the suspected cause?

  1. Detect fabricated citations despite polished formatting
  2. Check hidden assumptions behind inconsistent recommendations
  3. Find missing source context behind invented details
  4. Reproduce the error with and without the suspected earlier instructions

Explanation

Correct Answer

D:

Holding the alert text and model configuration constant while toggling the suspected harness instructions isolates one possible cause of the misroute. Compare the outputs before rewriting the live classification prompt.

Incorrect Answers

A:

That option is designed to verify the reality and contents of citations when invented evidence hidden by professional styling is the underlying concern. The decision described here calls for the team to test whether a harness instruction causes the misrouting to handle an integration-specific routing defect of uncertain origin.

B:

This selection provides a path to make decision priorities explicit and consistent, addressing recommendations varying with unstated trade-offs. It would not fix the identified an integration-specific routing defect of uncertain origin; test whether a harness instruction causes the misrouting is the missing step.

C:

Missing evidence, not poor wording style, is the likely cause of unsupported details; that principle matters when the concern is a response built without necessary evidence. Here, the issue of an integration-specific routing defect of uncertain origin still requires a way to test whether a harness instruction causes the misrouting.

Question 87

A public-benefits agency uses Claude to draft rejection notices. Reviewers find that some letters insult applicants, and applicants cannot correct source-record mistakes before a denial becomes final. Which TWO safeguards directly address these defects? (Choose TWO.)

  1. Avoid invented statements from real individuals
  2. Describe rejected applicants without derogatory labels
  3. Let applicants correct records and appeal denials
  4. Assess environmental and resource costs proportionately

Explanation

Correct Answers

B:

Describe the documented event and applicable criterion without attacking the applicant’s character. This keeps automated correspondence respectful and reviewable.

C:

Applicants need a meaningful opportunity to contest incorrect records and seek accountable reconsideration. A model-generated ranking must not close that route.

Incorrect Answers

A:

The benefit is tied to false personal attribution in generated copy; the method seeks to preserve authenticity of attributed speech. In this situation, the priority is to use neutral descriptions and establish a meaningful appeal procedure.

D:

Even if the team could match computation to real task value, it would still confront derogatory rejection wording and the absence of a correction route. The proposed step does not fulfill the separate requirement to use neutral descriptions and establish a meaningful appeal procedure.

Question 88

A Project tells Claude to "be excellent and responsible" but gives no testable directions for sources, citations, or handling missing data. What is the most appropriate next action?

  1. Make instructions actionable rather than aspirational
  2. Provide an exception rule for conflicting requirements
  3. Differentiate mandatory rules from optional preferences
  4. Separate quoted source text from privileged directions

Explanation

Correct Answer

A:

Concrete behaviors such as citing supplied evidence, marking unknown fields, and requesting review are verifiable. General virtues alone cannot guide a repeatable process.

Incorrect Answers

B:

There is a legitimate reason to resolve conflicts using explicit mandatory exceptions when no rule for competing standing directions is the barrier. For the reported vague high-level instructions without operational rules, the team needs to turn abstract values into observable behaviors instead.

C:

One could justify the intervention in a situation involving conflicts between essential rules and style preferences, since it aims to order instructions by operational criticality. It does not correct vague high-level instructions without operational rules in this scenario.

D:

This approach would be meaningful for malicious directives embedded in retrieved content: standing instructions should treat retrieved content as data rather than higher-authority directions. It does not directly resolve vague high-level instructions without operational rules, which requires the team to turn abstract values into observable behaviors.

Question 89

A department asks for "an AI workflow" without explaining the service delay or customer result it hopes to improve. Which intervention is best suited to this situation?

  1. Identify the business outcome before selecting a tool
  2. Capture data classification before deciding architecture
  3. Account for accessibility and language needs
  4. List exception paths alongside normal flows
  5. Document dependencies on existing systems

Explanation

Correct Answer

A:

A usable requirement starts with a measurable business problem and its owner. Product selection can follow once the desired outcome and constraints are clear.

Incorrect Answers

B:

It would be a useful change if sensitive source material ignored during design were present: the goal there is to classify inputs and permitted handling paths. The record describes a technology request without an operational goal instead.

C:

The benefit is tied to excluded users in the initial requirements; the method seeks to capture diverse user access requirements. In this situation, the priority is to define the outcome and owning process.

D:

The foil relates to requirements limited to the happy path, for which the goal is to capture nonstandard cases and escalation routes. The team must instead define the outcome and owning process because the actual weakness is a technology request without an operational goal.

E:

This choice may produce the benefit of being able to identify system interfaces and ownership. That does not remove a technology request without an operational goal; the work calls for a means to define the outcome and owning process.

Question 90

An adoption pilot reports impressive draft quality but does not track usage cost per completed and accepted task. Which approach best meets the stated need?

  1. Use observable failure rates to guide upgrades
  2. Pilot the decision workflow before broad rollout
  3. Keep a human gate for consequential outputs
  4. Record per-task cost and quality together

Explanation

Correct Answer

D:

Cost per accepted result incorporates both model expense and correction overhead. It is more useful than comparing token prices without measuring final business outcomes.

Incorrect Answers

A:

Solving model changes driven by isolated failures would mean arranging to upgrade only where systematic errors justify it. The present evidence describes unclear return on model usage, so the action does not directly evaluate cost per approved task.

B:

The idea behind the choice is that a controlled pilot tests representative tasks, failure modes, and support needs first. However, that relates to purchasing decisions made without workload data, not the stated requirement to evaluate cost per approved task.

C:

It would contribute toward incorporate risk-based human validation, which is useful for high-impact errors despite acceptable average accuracy. The scenario requires a response to unclear return on model usage rather than that separate improvement.

Claude Certified Associate - Foundations (CCAO-F) Practice Tests