Google Professional Cloud Security Engineer Question 366
Select 4Google Cloud PlatformYour company is deploying an AI/ML system on Google Cloud to analyze sensitive customer data, including personally identifiable information (PII). To protect against unintentional exploitation of data or model weaknesses, which actions should you take as part of implementing security and privacy controls?
- A
Implement data anonymization techniques, such as differential privacy, before training the model.
- B
Restrict model access to only authorized users using IAM roles and permissions.
- C
Disable the use of encryption for the model's training data to optimize performance.
- D
Perform adversarial testing to identify vulnerabilities in the AI/ML model.
- E
Use Vertex AI’s built-in Data Loss Prevention (DLP) features to detect and mask sensitive data during preprocessing.
Show answer and explanation
Correct answers: A, B, D, E
Explanation
To secure AI/ML systems and protect against exploitation, it’s crucial to implement data anonymization, restrict access to the model, and ensure sensitive information is safeguarded using tools like DLP. Additionally, adversarial testing is necessary to identify and address potential vulnerabilities in the model. Avoid actions that compromise security, such as disabling encryption for performance reasons.
- A. Correct.
Data anonymization techniques like differential privacy are critical to protecting sensitive data during model training and preventing unintentional exploitation.
- B. Correct.
Restricting access to the model ensures that only authorized users can interact with it, reducing the risk of unauthorized access or manipulation.
- C. Incorrect.
Disabling encryption for training data exposes sensitive information and creates a major security vulnerability. Encryption is essential for protecting data at rest and in transit.
- D. Correct.
Adversarial testing helps identify weaknesses in the model that could be exploited, such as susceptibility to adversarial attacks or model inversion.
- E. Correct.
Vertex AI’s DLP features are designed to help identify and mask sensitive data during preprocessing, ensuring compliance with data privacy requirements and reducing risk.