CDPSE Privacy Engineering Practice Test 5 — 15 Questions

Question 1

Which THREE privacy engineering activities are performed during the 'Maintenance' phase of the system lifecycle?

Accepted Answer

Patching security vulnerabilities. In the Maintenance phase of the system lifecycle, privacy engineering focuses on keeping the deployed system compliant and secure over time. Option B (Patching security vulnerabilities) is correct because applying security patches during maintenance prevents newly discovered exploits from compromising personal data that the system already processes. Option C (Continuous privacy monitoring) is correct because ongoing monitoring detects drift, unauthorized data flows, and emerging privacy risks in a live system, which is a core maintenance activity. Option D (Periodic DPIA review) is correct because a Data Protection Impact Assessment must be revisited when processing changes or new risks arise, and this recurring review belongs to the maintenance phase. Option A (Initial design wireframing) is not correct because wireframing happens during the design phase, before the system is built. Option E (Project budget estimation) is not correct because budgeting is a project management activity typically performed during initiation or planning, not a privacy engineering task in maintenance.

Answer

Initial design wireframing

Answer

Project budget estimation

Question 2

A privacy engineer is implementing differential privacy in a data analytics pipeline. To ensure the privacy budget remains intact over multiple queries, which technique should be applied?

Accepted Answer

Utilizing composition theorems to track cumulative privacy loss. In differential privacy, each query consumes a portion of the privacy budget (epsilon). Composition theorems provide a formal way to track cumulative privacy loss across multiple queries, ensuring the total budget is not exceeded. This is essential for maintaining the promised privacy guarantee over a series of analyses.

Answer

Implementing k-anonymity on the original dataset

Answer

Increasing the sensitivity of the query function

Answer

Applying a fixed noise distribution regardless of query count

Question 3

When pseudonymizing a database using SHA-256 hashing, what is the most critical requirement to prevent re-identification via brute-force or dictionary attacks?

Accepted Answer

Using a strong, unique salt. A strong, unique salt ensures that identical plaintext values produce different hashes and defeats precomputed rainbow tables and cross-record correlation. For low-entropy inputs like names, SSNs, or email addresses, an unsalted SHA-256 hash is trivially reversible by brute force or dictionary attack because the input space is small; a per-record random salt forces the attacker to attack each record independently, making mass re-identification computationally infeasible.

Answer

Encrypting the hashing algorithm

Answer

Reducing the input entropy

Answer

Truncating the output hash

Question 4

A web application stores user logs in a cleartext format. To implement 'Privacy by Design', which technical control is the most appropriate first step to minimize data exposure?

Accepted Answer

Apply data masking to the log fields. Data masking replaces or obscures sensitive values in log fields so that even if logs are accessed or leaked, the underlying personal data is not exposed. As a first step under Privacy by Design, it directly minimizes data exposure at the point of storage without requiring architectural changes. It is the most targeted control for cleartext logs.

Answer

Rotate encryption keys weekly

Answer

Implement full-disk encryption

Answer

Enable verbose logging for debugging

Question 5

You are integrating a homomorphic encryption library into a cloud-based financial system. What is the primary trade-off you must communicate to stakeholders regarding the implementation of Fully Homomorphic Encryption (FHE)?

Accepted Answer

Significant latency during computation. Fully Homomorphic Encryption (FHE) enables arbitrary computations on encrypted data, but the cryptographic operations are orders of magnitude slower than plaintext computations, causing significant latency. This latency is the primary trade-off that must be communicated to stakeholders, as it impacts system responsiveness and throughput. Other options like storage or integrity are not the main concerns.

Answer

Incompatibility with cloud storage

Answer

Increased storage requirements

Answer

Loss of data integrity

Question 6

Which feature in Microsoft Purview Information Protection should be used to automatically identify and classify documents containing credit card numbers as they are created?

Accepted Answer

Sensitivity labels with auto-labeling. Sensitivity labels with auto-labeling in Microsoft Purview Information Protection can automatically detect sensitive information types such as credit card numbers using built-in or custom sensitive information types (SITs) and apply classification labels at creation or modification. This directly fulfills the requirement to identify and classify documents containing credit card numbers automatically. It is the classification mechanism, not just a protective control.

Answer

eDiscovery search

Answer

Data Loss Prevention (DLP) policy

Answer

Information Rights Management (IRM)

Question 7

In a federated learning architecture, how is the model trained without exposing the underlying local datasets?

Accepted Answer

By sending locally computed gradients to a central server. In federated learning, each participant trains a local model on its own data and sends only the computed gradients (or model updates) to a central server, which aggregates them into a global model. The raw local datasets never leave the participant's environment, so training occurs without exposing the underlying data.

Answer

By using homomorphic encryption for every data point

Answer

By moving all datasets to a centralized data lake

Answer

By anonymizing the data using k-anonymity

Question 8

An organization uses a 'Privacy-preserving Synthetic Data' generator. What is the primary purpose of this tool in a development environment?

Accepted Answer

To enable testing without using actual production PII. Privacy-preserving synthetic data generators produce artificial datasets that statistically mirror the structure and distributions of real production data without containing any actual PII. This lets developers and QA teams build, test, and train models against realistic data shapes while eliminating the regulatory and security risks of exposing live personal data in non-production environments. The primary purpose is therefore enabling testing without using actual production PII.

Answer

To replace the need for database backups

Answer

To verify the identity of developers

Answer

To increase the accuracy of machine learning models

Question 9

When configuring 'Dynamic Data Masking' in SQL Server, which permission is required for a user to see the unmasked data?

Accepted Answer

UNMASK permission. In SQL Server Dynamic Data Masking (DDM), users who need to see the actual unmasked values must be granted the UNMASK permission, either at the database level (GRANT UNMASK TO user) or via a role that includes it. By default, users with SELECT permission see masked values for columns defined with a masking function; UNMASK is the explicit override that reveals the underlying data. This is the specific permission the question is asking about.

Answer

db_datareader

Answer

SELECT permission

Answer

sysadmin role

Question 10

A privacy engineer is implementing 'Tokenization' for credit card processing. Where should the 'vault' be situated to ensure the highest level of privacy?

Accepted Answer

In a physically or logically segregated environment. Tokenization replaces sensitive data with non-sensitive tokens, and the mapping between tokens and original values is held in a token vault. To maximize privacy, the vault must be isolated in a physically or logically segregated environment so that compromise of the application, web server, or logs does not expose the token-to-data mapping. Segregation enforces defense in depth and limits blast radius if any front-end component is breached.

Answer

In the same database as the application logs

Answer

On the end-user's local workstation

Answer

Integrated directly into the web server memory

Question 11

Which of the following is the most effective technical control for implementing the 'Right to be Forgotten' in a distributed microservices environment?

Accepted Answer

Implementing a centralized 'Identity Deletion' event bus. In a distributed microservices environment, personal data is scattered across many services and data stores, so honoring erasure requires coordinated action. A centralized 'Identity Deletion' event bus publishes a deletion event that each subscribing service consumes to erase or anonymize its copy of the subject's data, providing a scalable, auditable, and consistent mechanism. This event-driven pattern ensures no service is missed and creates a traceable record of propagation.

Answer

Deleting individual database records in each service

Answer

Archiving logs for 7 years

Answer

Increasing database logging levels

Question 12

When using 'k-anonymity' to protect a dataset, what does the parameter 'k' represent?

Accepted Answer

The minimum number of records sharing the same quasi-identifiers. In k-anonymity, k is the minimum group size such that every record is indistinguishable from at least k-1 other records with respect to the quasi-identifier set (age, ZIP, gender, etc.). A dataset satisfies k-anonymity if each combination of quasi-identifier values appears at least k times, preventing re-identification by linking to external data. Higher k means stronger anonymity but greater information loss.

Answer

The number of attributes removed

Answer

The level of noise added to the data

Answer

The total number of rows in the dataset

Question 13

To ensure that data access logs cannot be tampered with by an administrator with high privileges, which technical control should be implemented?

Accepted Answer

Immutable log storage. Immutable log storage, such as WORM (Write Once Read Many) storage or blockchain-backed logs, ensures that even a privileged administrator cannot alter or delete access logs after they are written. This provides tamper-evidence and tamper-resistance, which is essential for audit integrity. RBAC, encryption at rest, and MFA do not prevent a high-privileged administrator from modifying logs.

Answer

Role-Based Access Control (RBAC)

Answer

Database encryption at rest

Answer

Multi-factor authentication (MFA)

Question 14

What is the primary privacy advantage of using 'On-Device Processing' for machine learning inference?

Accepted Answer

Reduction in data transfer of sensitive inputs. On-device inference runs the ML model locally on the user's device (phone, laptop, edge sensor), so raw sensitive inputs such as voice, images, or keystrokes never leave the device boundary. This eliminates the data transfer of personal data to a cloud endpoint, which is the core privacy advantage — it reduces the attack surface, avoids third-party data exposure, and supports data minimization principles under GDPR/CCPA. The model weights may still be downloaded, but the personal data itself stays local.

Answer

Easier debugging of the model

Answer

Faster cloud infrastructure scaling

Answer

Improved model accuracy

Question 15

You are configuring an API gateway to implement 'Data Minimization' via response filtering. How is this typically achieved for JSON payloads?

Accepted Answer

By implementing field-level response transformation. Data minimization in an API gateway response is achieved by transforming the response payload to include only the fields the client actually needs, which for JSON means field-level response transformation (e.g., using mapping templates, JSON schema filters, or response models). This strips out unnecessary or sensitive attributes before they leave the gateway, enforcing minimization at the boundary. It is a payload-shaping control, not a transport or backend control.

Answer

By using TLS 1.3 encryption

Answer

By increasing the request timeout

Answer

By modifying the backend database schema