Practice Exams:

ISC CCSP Dumps

EXAM: Certified Cloud Security Professional (CCSP)
Verified By Experts
Premium Bundle
$40.00
$59.97 $99.97

CCSP Premium Bundle


  • Premium File 512 Questions & Answers
  • Training Course 43 Lectures
  • Study Guide 571 Pages

ISC CCSP Practice Test 33

Question 320

A medical model team imports partner image data whose authorization and origin cannot be traced. Suppose the team makes this shortcut instead: Mix unvetted external training examples into a trusted dataset because more records must improve the model. Which outcome is the most likely direct consequence?

  1. Unauthorized declassification bypasses access controls and retention obligations
  2. Copies spread beyond the original access boundary and cannot be centrally restricted
  3. Decision makers rely on an unachievable containment guarantee after confidential content was accessed
  4. A manipulated training source changes cloud decisions and undermines data privacy promises

Answer & Explanations

Correct answer: D

A.

This failure concerns another scenario: A research analyst wants to downgrade partner-provided confidential datasets for public collaboration. In the stated case: A medical model team imports partner image data whose authorization and origin cannot be traced. The relevant consequence is: A manipulated training source changes cloud decisions and undermines data privacy promises. The control that prevents it is: Establish signed dataset versions, provenance checks and validation thresholds before using third-party training samples.

B.

This failure concerns another scenario: An investment firm distributes sensitive forecasts to temporary advisers and needs access to end after the engagement. In the stated case: A medical model team imports partner image data whose authorization and origin cannot be traced. The relevant consequence is: A manipulated training source changes cloud decisions and undermines data privacy promises. The control that prevents it is: Establish signed dataset versions, provenance checks and validation thresholds before using third-party training samples.

C.

This failure concerns another scenario: A compliance officer revokes a departing contractor's protected files after the contractor previously printed a copy. In the stated case: A medical model team imports partner image data whose authorization and origin cannot be traced. The relevant consequence is: A manipulated training source changes cloud decisions and undermines data privacy promises. The control that prevents it is: Establish signed dataset versions, provenance checks and validation thresholds before using third-party training samples.

D.

A manipulated training source changes cloud decisions and undermines data privacy promises. This is the risk in the stated situation: A medical model team imports partner image data whose authorization and origin cannot be traced. The risky shortcut is: Mix unvetted external training examples into a trusted dataset because more records must improve the model. Prevent it by: Establish signed dataset versions, provenance checks and validation thresholds before using third-party training samples.

Question 321

A hospital deploys a predictive model that repeats fragments from a small confidential training corpus. Which cloud security control or governance decision most directly addresses the described issue while keeping the change scoped?

  1. Apply privacy engineering such as minimum necessary attributes, controlled access and validated aggregation or privacy-preserving training where appropriate
  2. Document ingress, transformation, service-to-service transfer and output sinks for regulated data using validated flow evidence
  3. Enforce private object access, deny-public guardrails and least-privilege bucket policies with explicit negative tests
  4. Place persistent state in approved encrypted durable volumes or object stores while treating container and ephemeral disks as disposable

Answer & Explanations

Correct answer: A

A.

Apply privacy engineering such as minimum necessary attributes, controlled access and validated aggregation or privacy-preserving training where appropriate. The technical reason is: Models can reveal information about training members through outputs even if raw datasets are never publicly exposed. The stated condition is: A hospital deploys a predictive model that repeats fragments from a small confidential training corpus. Confirm the result by: Probe a nonproduction model for memorized identifiers and check mitigation results against utility and privacy targets.

B.

This action is appropriate in a different case: A payment analytics pipeline stores records in one jurisdiction but exports detailed traces to a global observability tenant. There, the reason is: Data security and residency controls depend on the complete processing path, not only the primary storage region. The current case is: A hospital deploys a predictive model that repeats fragments from a small confidential training corpus. The appropriate response is: Apply privacy engineering such as minimum necessary attributes, controlled access and validated aggregation or privacy-preserving training where appropriate.

C.

This action is appropriate in a different case: A research institute stores unpublished genomic samples in a cloud object bucket with an accidental public-read setting. There, the reason is: Object URLs and bucket names are not authorization controls and inherited ACLs can expose entire datasets. The current case is: A hospital deploys a predictive model that repeats fragments from a small confidential training corpus. The appropriate response is: Apply privacy engineering such as minimum necessary attributes, controlled access and validated aggregation or privacy-preserving training where appropriate.

D.

This action is appropriate in a different case: A financial microservice writes transactions to local container scratch space during nightly autoscaling. There, the reason is: Ephemeral compute storage can disappear during restarts and is not a substitute for durable recovery points. The current case is: A hospital deploys a predictive model that repeats fragments from a small confidential training corpus. The appropriate response is: Apply privacy engineering such as minimum necessary attributes, controlled access and validated aggregation or privacy-preserving training where appropriate.

Question 322

A hospital deploys a predictive model that repeats fragments from a small confidential training corpus. An engineer recommends: Apply privacy engineering such as minimum necessary attributes, controlled access and validated aggregation or privacy-preserving training where appropriate. Which technical explanation most directly justifies this recommendation?

  1. Ephemeral compute storage can disappear during restarts and is not a substitute for durable recovery points
  2. Models can reveal information about training members through outputs even if raw datasets are never publicly exposed
  3. A data encryption key protects content while a separate key-encryption key provides centralized wrapping and revocation
  4. Substituting controlled tokens reduces direct identifier exposure while DLP detects restricted content leaving authorized boundaries

Answer & Explanations

Correct answer: B

A.

This principle explains a different technical intervention: Place persistent state in approved encrypted durable volumes or object stores while treating container and ephemeral disks as disposable. It applies to: A financial microservice writes transactions to local container scratch space during nightly autoscaling. The actual case is: A hospital deploys a predictive model that repeats fragments from a small confidential training corpus. The mechanism relevant here is: Models can reveal information about training members through outputs even if raw datasets are never publicly exposed.

B.

Models can reveal information about training members through outputs even if raw datasets are never publicly exposed. That mechanism explains the decision: Apply privacy engineering such as minimum necessary attributes, controlled access and validated aggregation or privacy-preserving training where appropriate. In the reported situation: A hospital deploys a predictive model that repeats fragments from a small confidential training corpus. Confirmation requires: Probe a nonproduction model for memorized identifiers and check mitigation results against utility and privacy targets.

C.

This principle explains a different technical intervention: Use envelope encryption with monitored tenant-controlled master key permissions for sensitive cloud records. It applies to: An insurer wants to restrict decryption of archived files without re-encrypting every object manually. The actual case is: A hospital deploys a predictive model that repeats fragments from a small confidential training corpus. The mechanism relevant here is: Models can reveal information about training members through outputs even if raw datasets are never publicly exposed.

D.

This principle explains a different technical intervention: Tokenize sensitive payment identifiers and enforce content-aware DLP checks at approved cloud egress paths. It applies to: A retailer shares customer transaction analytics externally and needs to prevent exposure of underlying card numbers. The actual case is: A hospital deploys a predictive model that repeats fragments from a small confidential training corpus. The mechanism relevant here is: Models can reveal information about training members through outputs even if raw datasets are never publicly exposed.

Question 323

A hospital deploys a predictive model that repeats fragments from a small confidential training corpus. After applying the relevant fix, which observable evidence would most convincingly establish that the underlying issue has been resolved?

  1. Export a test record through the analytics pipeline and confirm tokenized values persist and raw account numbers are blocked
  2. Seed test documents containing labeled sensitive terms and confirm discovery covers intended stores with acceptable false positives
  3. Probe a nonproduction model for memorized identifiers and check mitigation results against utility and privacy targets
  4. Trace known synthetic identifiers across SQL tables, streamed JSON and downstream warehouse views

Answer & Explanations

Correct answer: C

A.

This observation would verify a different remedy: Tokenize sensitive payment identifiers and enforce content-aware DLP checks at approved cloud egress paths. That remedy applies to: A retailer shares customer transaction analytics externally and needs to prevent exposure of underlying card numbers. The present condition is: A hospital deploys a predictive model that repeats fragments from a small confidential training corpus. To assess it, check: Probe a nonproduction model for memorized identifiers and check mitigation results against utility and privacy targets.

B.

This observation would verify a different remedy: Run authorized content discovery across object documents and free-text files before assigning controls. That remedy applies to: A legal consultancy migrates years of shared files whose contents were never classified. The present condition is: A hospital deploys a predictive model that repeats fragments from a small confidential training corpus. To assess it, check: Probe a nonproduction model for memorized identifiers and check mitigation results against utility and privacy targets.

C.

Probe a nonproduction model for memorized identifiers and check mitigation results against utility and privacy targets. This check directly tests: Apply privacy engineering such as minimum necessary attributes, controlled access and validated aggregation or privacy-preserving training where appropriate. The underlying behavior is: Models can reveal information about training members through outputs even if raw datasets are never publicly exposed. The case is: A hospital deploys a predictive model that repeats fragments from a small confidential training corpus.

D.

This observation would verify a different remedy: Combine schema inspection, metadata scanning and controlled sampling to discover sensitive data in managed databases and JSON event streams. That remedy applies to: A delivery app stores customer addresses in SQL but sends full addresses in semi-structured diagnostic events. The present condition is: A hospital deploys a predictive model that repeats fragments from a small confidential training corpus. To assess it, check: Probe a nonproduction model for memorized identifiers and check mitigation results against utility and privacy targets.

Question 324

A hospital deploys a predictive model that repeats fragments from a small confidential training corpus. Which apparently convenient shortcut would most directly undermine safe resolution of this specific situation?

  1. Scan only the production relational database and conclude every customer attribute has been found
  2. Mark every new bucket public by default until somebody reports a sensitive file
  3. Let a developer remove a confidential tag to accelerate data sharing without business approval
  4. Assume a trained model cannot reveal personal data because it no longer stores the original SQL table

Answer & Explanations

Correct answer: D

A.

This proposed shortcut concerns a separate case: A delivery app stores customer addresses in SQL but sends full addresses in semi-structured diagnostic events. The risk in this question concerns: A hospital deploys a predictive model that repeats fragments from a small confidential training corpus. Here the unsafe shortcut is: Assume a trained model cannot reveal personal data because it no longer stores the original SQL table. Use this response instead: Apply privacy engineering such as minimum necessary attributes, controlled access and validated aggregation or privacy-preserving training where appropriate.

B.

This proposed shortcut concerns a separate case: A biomedical lab creates datasets faster than reviewers can apply restricted data labels manually. The risk in this question concerns: A hospital deploys a predictive model that repeats fragments from a small confidential training corpus. Here the unsafe shortcut is: Assume a trained model cannot reveal personal data because it no longer stores the original SQL table. Use this response instead: Apply privacy engineering such as minimum necessary attributes, controlled access and validated aggregation or privacy-preserving training where appropriate.

C.

This proposed shortcut concerns a separate case: A research analyst wants to downgrade partner-provided confidential datasets for public collaboration. The risk in this question concerns: A hospital deploys a predictive model that repeats fragments from a small confidential training corpus. Here the unsafe shortcut is: Assume a trained model cannot reveal personal data because it no longer stores the original SQL table. Use this response instead: Apply privacy engineering such as minimum necessary attributes, controlled access and validated aggregation or privacy-preserving training where appropriate.

D.

Assume a trained model cannot reveal personal data because it no longer stores the original SQL table. This shortcut is unsafe for: A hospital deploys a predictive model that repeats fragments from a small confidential training corpus. The resulting failure can be: An attacker infers participation or extracts sensitive training attributes from repeated model queries. The safer response is: Apply privacy engineering such as minimum necessary attributes, controlled access and validated aggregation or privacy-preserving training where appropriate.

Question 325

A hospital deploys a predictive model that repeats fragments from a small confidential training corpus. Suppose the team makes this shortcut instead: Assume a trained model cannot reveal personal data because it no longer stores the original SQL table. Which outcome is the most likely direct consequence?

  1. An attacker infers participation or extracts sensitive training attributes from repeated model queries
  2. Unauthorized declassification bypasses access controls and retention obligations
  3. Copies spread beyond the original access boundary and cannot be centrally restricted
  4. Decision makers rely on an unachievable containment guarantee after confidential content was accessed

Answer & Explanations

Correct answer: A

A.

An attacker infers participation or extracts sensitive training attributes from repeated model queries. This is the risk in the stated situation: A hospital deploys a predictive model that repeats fragments from a small confidential training corpus. The risky shortcut is: Assume a trained model cannot reveal personal data because it no longer stores the original SQL table. Prevent it by: Apply privacy engineering such as minimum necessary attributes, controlled access and validated aggregation or privacy-preserving training where appropriate.

B.

This failure concerns another scenario: A research analyst wants to downgrade partner-provided confidential datasets for public collaboration. In the stated case: A hospital deploys a predictive model that repeats fragments from a small confidential training corpus. The relevant consequence is: An attacker infers participation or extracts sensitive training attributes from repeated model queries. The control that prevents it is: Apply privacy engineering such as minimum necessary attributes, controlled access and validated aggregation or privacy-preserving training where appropriate.

C.

This failure concerns another scenario: An investment firm distributes sensitive forecasts to temporary advisers and needs access to end after the engagement. In the stated case: A hospital deploys a predictive model that repeats fragments from a small confidential training corpus. The relevant consequence is: An attacker infers participation or extracts sensitive training attributes from repeated model queries. The control that prevents it is: Apply privacy engineering such as minimum necessary attributes, controlled access and validated aggregation or privacy-preserving training where appropriate.

D.

This failure concerns another scenario: A compliance officer revokes a departing contractor's protected files after the contractor previously printed a copy. In the stated case: A hospital deploys a predictive model that repeats fragments from a small confidential training corpus. The relevant consequence is: An attacker infers participation or extracts sensitive training attributes from repeated model queries. The control that prevents it is: Apply privacy engineering such as minimum necessary attributes, controlled access and validated aggregation or privacy-preserving training where appropriate.

Question 326

An insurer exposes a model API where repeated queries can reveal whether a specific claimant was in training data. Which cloud security control or governance decision most directly addresses the described issue while keeping the change scoped?

  1. Place persistent state in approved encrypted durable volumes or object stores while treating container and ephemeral disks as disposable
  2. Apply privacy engineering such as minimum necessary attributes, controlled access and validated aggregation or privacy-preserving training where appropriate
  3. Use envelope encryption with monitored tenant-controlled master key permissions for sensitive cloud records
  4. Tokenize sensitive payment identifiers and enforce content-aware DLP checks at approved cloud egress paths

Answer & Explanations

Correct answer: B

A.

This action is appropriate in a different case: A financial microservice writes transactions to local container scratch space during nightly autoscaling. There, the reason is: Ephemeral compute storage can disappear during restarts and is not a substitute for durable recovery points. The current case is: An insurer exposes a model API where repeated queries can reveal whether a specific claimant was in training data. The appropriate response is: Apply privacy engineering such as minimum necessary attributes, controlled access and validated aggregation or privacy-preserving training where appropriate.

B.

Apply privacy engineering such as minimum necessary attributes, controlled access and validated aggregation or privacy-preserving training where appropriate. The technical reason is: Models can reveal information about training members through outputs even if raw datasets are never publicly exposed. The stated condition is: An insurer exposes a model API where repeated queries can reveal whether a specific claimant was in training data. Confirm the result by: Probe a nonproduction model for memorized identifiers and check mitigation results against utility and privacy targets.

C.

This action is appropriate in a different case: An insurer wants to restrict decryption of archived files without re-encrypting every object manually. There, the reason is: A data encryption key protects content while a separate key-encryption key provides centralized wrapping and revocation. The current case is: An insurer exposes a model API where repeated queries can reveal whether a specific claimant was in training data. The appropriate response is: Apply privacy engineering such as minimum necessary attributes, controlled access and validated aggregation or privacy-preserving training where appropriate.

D.

This action is appropriate in a different case: A retailer shares customer transaction analytics externally and needs to prevent exposure of underlying card numbers. There, the reason is: Substituting controlled tokens reduces direct identifier exposure while DLP detects restricted content leaving authorized boundaries. The current case is: An insurer exposes a model API where repeated queries can reveal whether a specific claimant was in training data. The appropriate response is: Apply privacy engineering such as minimum necessary attributes, controlled access and validated aggregation or privacy-preserving training where appropriate.

Question 327

An insurer exposes a model API where repeated queries can reveal whether a specific claimant was in training data. An engineer recommends: Apply privacy engineering such as minimum necessary attributes, controlled access and validated aggregation or privacy-preserving training where appropriate. Which technical explanation most directly justifies this recommendation?

  1. Substituting controlled tokens reduces direct identifier exposure while DLP detects restricted content leaving authorized boundaries
  2. File extensions and container names do not establish whether unstructured files hold personal or regulated content
  3. Models can reveal information about training members through outputs even if raw datasets are never publicly exposed
  4. Sensitive fields can appear in both typed columns and unstructured event payloads that a database-only inventory misses

Answer & Explanations

Correct answer: C

A.

This principle explains a different technical intervention: Tokenize sensitive payment identifiers and enforce content-aware DLP checks at approved cloud egress paths. It applies to: A retailer shares customer transaction analytics externally and needs to prevent exposure of underlying card numbers. The actual case is: An insurer exposes a model API where repeated queries can reveal whether a specific claimant was in training data. The mechanism relevant here is: Models can reveal information about training members through outputs even if raw datasets are never publicly exposed.

B.

This principle explains a different technical intervention: Run authorized content discovery across object documents and free-text files before assigning controls. It applies to: A legal consultancy migrates years of shared files whose contents were never classified. The actual case is: An insurer exposes a model API where repeated queries can reveal whether a specific claimant was in training data. The mechanism relevant here is: Models can reveal information about training members through outputs even if raw datasets are never publicly exposed.

C.

Models can reveal information about training members through outputs even if raw datasets are never publicly exposed. That mechanism explains the decision: Apply privacy engineering such as minimum necessary attributes, controlled access and validated aggregation or privacy-preserving training where appropriate. In the reported situation: An insurer exposes a model API where repeated queries can reveal whether a specific claimant was in training data. Confirmation requires: Probe a nonproduction model for memorized identifiers and check mitigation results against utility and privacy targets.

D.

This principle explains a different technical intervention: Combine schema inspection, metadata scanning and controlled sampling to discover sensitive data in managed databases and JSON event streams. It applies to: A delivery app stores customer addresses in SQL but sends full addresses in semi-structured diagnostic events. The actual case is: An insurer exposes a model API where repeated queries can reveal whether a specific claimant was in training data. The mechanism relevant here is: Models can reveal information about training members through outputs even if raw datasets are never publicly exposed.

Question 328

An insurer exposes a model API where repeated queries can reveal whether a specific claimant was in training data. After applying the relevant fix, which observable evidence would most convincingly establish that the underlying issue has been resolved?

  1. Trace known synthetic identifiers across SQL tables, streamed JSON and downstream warehouse views
  2. Create synthetic confidential and public records and verify labels route each to the correct data-control policy
  3. Review a sample downgrade audit and confirm owner authorization, justifying evidence and downstream policy change
  4. Probe a nonproduction model for memorized identifiers and check mitigation results against utility and privacy targets

Answer & Explanations

Correct answer: D

A.

This observation would verify a different remedy: Combine schema inspection, metadata scanning and controlled sampling to discover sensitive data in managed databases and JSON event streams. That remedy applies to: A delivery app stores customer addresses in SQL but sends full addresses in semi-structured diagnostic events. The present condition is: An insurer exposes a model API where repeated queries can reveal whether a specific claimant was in training data. To assess it, check: Probe a nonproduction model for memorized identifiers and check mitigation results against utility and privacy targets.

B.

This observation would verify a different remedy: Assign cloud resource and data labels using approved classification rules with owner review for uncertain matches. That remedy applies to: A biomedical lab creates datasets faster than reviewers can apply restricted data labels manually. The present condition is: An insurer exposes a model API where repeated queries can reveal whether a specific claimant was in training data. To assess it, check: Probe a nonproduction model for memorized identifiers and check mitigation results against utility and privacy targets.

C.

This observation would verify a different remedy: Require accountable business owners to approve classification downgrades and retain reasons for material changes. That remedy applies to: A research analyst wants to downgrade partner-provided confidential datasets for public collaboration. The present condition is: An insurer exposes a model API where repeated queries can reveal whether a specific claimant was in training data. To assess it, check: Probe a nonproduction model for memorized identifiers and check mitigation results against utility and privacy targets.

D.

Probe a nonproduction model for memorized identifiers and check mitigation results against utility and privacy targets. This check directly tests: Apply privacy engineering such as minimum necessary attributes, controlled access and validated aggregation or privacy-preserving training where appropriate. The underlying behavior is: Models can reveal information about training members through outputs even if raw datasets are never publicly exposed. The case is: An insurer exposes a model API where repeated queries can reveal whether a specific claimant was in training data.

Question 329

An insurer exposes a model API where repeated queries can reveal whether a specific claimant was in training data. Which apparently convenient shortcut would most directly undermine safe resolution of this specific situation?

  1. Assume a trained model cannot reveal personal data because it no longer stores the original SQL table
  2. Let a developer remove a confidential tag to accelerate data sharing without business approval
  3. Email a confidential spreadsheet to every contractor as an unrestricted attachment and expect bucket permissions to control it
  4. Claim that revoking a certificate remotely deletes every screenshot and printout already made

Answer & Explanations

Correct answer: A

A.

Assume a trained model cannot reveal personal data because it no longer stores the original SQL table. This shortcut is unsafe for: An insurer exposes a model API where repeated queries can reveal whether a specific claimant was in training data. The resulting failure can be: An attacker infers participation or extracts sensitive training attributes from repeated model queries. The safer response is: Apply privacy engineering such as minimum necessary attributes, controlled access and validated aggregation or privacy-preserving training where appropriate.

B.

This proposed shortcut concerns a separate case: A research analyst wants to downgrade partner-provided confidential datasets for public collaboration. The risk in this question concerns: An insurer exposes a model API where repeated queries can reveal whether a specific claimant was in training data. Here the unsafe shortcut is: Assume a trained model cannot reveal personal data because it no longer stores the original SQL table. Use this response instead: Apply privacy engineering such as minimum necessary attributes, controlled access and validated aggregation or privacy-preserving training where appropriate.

C.

This proposed shortcut concerns a separate case: An investment firm distributes sensitive forecasts to temporary advisers and needs access to end after the engagement. The risk in this question concerns: An insurer exposes a model API where repeated queries can reveal whether a specific claimant was in training data. Here the unsafe shortcut is: Assume a trained model cannot reveal personal data because it no longer stores the original SQL table. Use this response instead: Apply privacy engineering such as minimum necessary attributes, controlled access and validated aggregation or privacy-preserving training where appropriate.

D.

This proposed shortcut concerns a separate case: A compliance officer revokes a departing contractor's protected files after the contractor previously printed a copy. The risk in this question concerns: An insurer exposes a model API where repeated queries can reveal whether a specific claimant was in training data. Here the unsafe shortcut is: Assume a trained model cannot reveal personal data because it no longer stores the original SQL table. Use this response instead: Apply privacy engineering such as minimum necessary attributes, controlled access and validated aggregation or privacy-preserving training where appropriate.

Question 330

An insurer exposes a model API where repeated queries can reveal whether a specific claimant was in training data. Suppose the team makes this shortcut instead: Assume a trained model cannot reveal personal data because it no longer stores the original SQL table. Which outcome is the most likely direct consequence?

  1. Decision makers rely on an unachievable containment guarantee after confidential content was accessed
  2. An attacker infers participation or extracts sensitive training attributes from repeated model queries
  3. Relevant records are destroyed before discovery or regulatory review and the organization loses defensible evidence
  4. Unnecessary historical data exposure and costs grow while regulated deletion requests cannot be honored

Answer & Explanations

Correct answer: B

A.

This failure concerns another scenario: A compliance officer revokes a departing contractor's protected files after the contractor previously printed a copy. In the stated case: An insurer exposes a model API where repeated queries can reveal whether a specific claimant was in training data. The relevant consequence is: An attacker infers participation or extracts sensitive training attributes from repeated model queries. The control that prevents it is: Apply privacy engineering such as minimum necessary attributes, controlled access and validated aggregation or privacy-preserving training where appropriate.

B.

An attacker infers participation or extracts sensitive training attributes from repeated model queries. This is the risk in the stated situation: An insurer exposes a model API where repeated queries can reveal whether a specific claimant was in training data. The risky shortcut is: Assume a trained model cannot reveal personal data because it no longer stores the original SQL table. Prevent it by: Apply privacy engineering such as minimum necessary attributes, controlled access and validated aggregation or privacy-preserving training where appropriate.

C.

This failure concerns another scenario: A cloud mail archive reaches normal deletion age while a regulator has requested specific correspondence. In the stated case: An insurer exposes a model API where repeated queries can reveal whether a specific claimant was in training data. The relevant consequence is: An attacker infers participation or extracts sensitive training attributes from repeated model queries. The control that prevents it is: Apply privacy engineering such as minimum necessary attributes, controlled access and validated aggregation or privacy-preserving training where appropriate.

D.

This failure concerns another scenario: A subscription provider stores every former customer backup indefinitely despite a documented disposal schedule. In the stated case: An insurer exposes a model API where repeated queries can reveal whether a specific claimant was in training data. The relevant consequence is: An attacker infers participation or extracts sensitive training attributes from repeated model queries. The control that prevents it is: Apply privacy engineering such as minimum necessary attributes, controlled access and validated aggregation or privacy-preserving training where appropriate.

Certified Cloud Security Professional (CCSP) (CCSP) Practice Tests