Google Professional Cloud Security Engineer Question 369
Select 3Google Cloud PlatformYour organization is deploying a machine learning model on Google Cloud AI Platform. The model uses sensitive financial data for training. How can you implement security and privacy controls to ensure that the data and model are not unintentionally exploited or leaked during the ML lifecycle?
- A
Use Data Loss Prevention (DLP) to de-identify sensitive data before training the model.
- B
Store the training dataset in a public Cloud Storage bucket to allow easy access for collaboration.
- C
Enable Customer-Managed Encryption Keys (CMEK) to encrypt model artifacts and training data.
- D
Implement Vertex AI Model Monitoring to detect and respond to data drift or anomalous predictions.
- E
Grant the 'Owner' role to all team members to ensure everyone has full access to the project.
Show answer and explanation
Correct answers: A, C, D
Explanation
To protect sensitive data and models in AI/ML systems, it is essential to implement de-identification techniques like DLP, use strong encryption mechanisms like CMEK, and employ monitoring tools like Vertex AI Model Monitoring. These measures help mitigate risks such as data breaches, model misuse, and unintentional exploitation. Avoid practices like over-permissive access roles or public data storage, which weaken security and privacy controls.
- A. Correct.
Using Data Loss Prevention (DLP) to de-identify sensitive data is a best practice for protecting sensitive information during the ML training process, ensuring privacy and compliance.
- B. Incorrect.
Storing training datasets in a public Cloud Storage bucket is insecure and could lead to unauthorized access to sensitive data, violating privacy and security requirements.
- C. Correct.
Enabling Customer-Managed Encryption Keys (CMEK) ensures that you maintain control over the encryption keys used to protect your data and model artifacts, enhancing security.
- D. Correct.
Vertex AI Model Monitoring allows you to monitor data drift and anomalous predictions, which helps identify potential exploitation or misuse of the model during its lifecycle.
- E. Incorrect.
Granting the 'Owner' role to all team members violates the principle of least privilege and increases the risk of unintentional or malicious exploitation of the project.