Data Governance and Generative AI

Google was one of the first in the industry to publish an AI/ML Privacy Commitment, which outlines our belief that customers should have the highest level of security and control over their data that is stored in the cloud. That commitment extends to Google Cloud generative AI products. Google helps ensure that its teams are following these commitments through robust data governance practices, which includes reviews of the data that Google Cloud uses in the development of its products. More details about how Google processes data can also be found in Google's Customer Data Processing Addendum (CDPA).

Foundation model development

By default, Google Cloud does not use Customer Data to train its foundation models as part of Google Cloud`s AI/ML Privacy Commitment.

Prompt Design

When a request is submitted with a prompt to our foundation model, Customer Data is encrypted in-transit and input to the foundation model to generate a response. Google processes Customer Data to provide the service, and does not use Customer Data to train its models without the express consent of its customers.

Model Tuning

Vertex AI offers a Parameter Efficient Tuning (PET) service that enables tuning of the model to specific tasks without having to rebuild the entire foundation model. Each tuning job results in the creation of a few additional learned parameters, called 'adapter weights'. The adapter weights are specific to the customer, and only available to the customer who tuned those weights. During inference, the foundation model receives the adapter weights, runs through the request and returns the results, without modifying the foundation model or storing the request.

Input data is Customer Data and is stored securely at every step along the way - encrypted at rest and in transit. Tuned weights are also stored securely, and Customers will have sole access to use any tuned models. The customer is able to control the encryption of stored adapters by using customer-managed encryption keys (CMEK), and can delete adapter weights at any time. Customer Data (e.g., prompts, input data) used to train adapter models will not be logged or used for improving the foundation model without the customer's permission.

What's next