Security, privacy, and compliance for Gemini in BigQuery

This document describes the controls that support the security of Gemini in BigQuery. These controls can also help you meet the privacy and regulatory requirements that apply to your business. Gemini in BigQuery is built on Google Cloud infrastructure. Your data remains in your control. For more information, see Service Specific Terms.

The following controls apply to Generally Available (GA) Gemini in BigQuery features:

  • Your data is not used for training models without your permission. Google does not use your prompts, responses, or schema information to train its models unless you explicitly opt in.
  • Your BigQuery data remains within your chosen location. Gemini in BigQuery respects your BigQuery data at rest data residency settings. The core BigQuery engine that runs queries and stores your data continues to honor your location constraints. For more information, see How Gemini in BigQuery processes data.
  • Gemini in BigQuery is covered by Google security and compliance offerings. Coverage includes certifications like SOC 1/2/3, ISO/IEC 27001, and HIPAA compliance. For more information, see Google security and compliance offerings.

The security, privacy, and compliance for Google Cloud services are a shared responsibility. Google secures the infrastructure that Google Cloud services run on, and it provides you with tools such as access controls to let you manage who has access to your services and resources. For more information about how the infrastructure is secured, see Google infrastructure security design overview.

Because Gemini is an evolving technology, it can generate output that's plausible-sounding but factually incorrect. We recommend that you validate all output from Gemini before you use it. For more information, see Gemini for Google Cloud and responsible AI.

Gemini in BigQuery architecture

The following diagram shows the components of the Gemini in BigQuery architecture.

Chart of Gemini in BigQuery customer data boundary and Google Cloud trust boundary.

How Gemini in BigQuery processes data

When a user uses Gemini in BigQuery, a prompt and relevant context are sent to Google's large language models (LLMs) for processing. Google manages the specific models used to generate Gemini in BigQuery responses.

  1. Prompt. A user enters a prompt as a natural language question, such as "Show me the top 5 customers by sales last quarter". Or, a user types a partial SQL or Python snippet in the Google Cloud console in BigQuery Studio with Gemini in BigQuery enabled.
  2. Contextualization. Gemini in BigQuery accesses the relevant metadata and schema of your BigQuery tables to add context to your prompt. Contextual information can include sampling data from tables and job histories. Gemini in BigQuery only has access to resources to which the user has access.
  3. Gemini processing. The prompt and contextual information are sent to Gemini's LLMs for processing. Gemini generates a response, such as a SQL query, a data insight, or a Python code snippet.
  4. Response. The response is returned to the BigQuery interface. The user can then run the generated code, modify it, or continue to iterate on the response by using Gemini. You can provide feedback from Gemini in BigQuery in the Google Cloud console. To learn more about providing feedback, see Provide feedback.

The following diagram compares normal SQL query execution starting from code execution to SQL query generation.

Division between global infrastructure and fully regionalized BigQuery resources.

Security controls

Gemini in BigQuery uses the security controls of Google Cloud to help protect your data and resources. These controls include the following:

  • Authentication. Users authenticate by using their Google Cloud credentials, which can be integrated with your existing identity provider.
  • Access controls. You can use Identity and Access Management (IAM) to control who has access to Gemini in BigQuery and what actions they can perform.
  • Network security and VPC-SC. Gemini in BigQuery traffic is encrypted in transit and at rest. You can also use VPC Service Controls to create a security-enhanced perimeter around your BigQuery resources.

Data protection and privacy

Gemini in BigQuery is designed to protect the privacy of your data. Google's privacy policies and commitments apply to all data processed by Gemini in BigQuery.

  • Data encryption. Your data is encrypted at rest and in transit.
  • Data access. Google personnel have limited and audited access to your data.
  • Data residency. Your BigQuery data at rest is stored and processed in the Google Cloud region you select. However, the Gemini large language model (LLM) processing of prompts and contextual metadata is a global service and doesn't adhere to data residency in use constraints.

Certifications and capabilities

Generally available (GA) Gemini in BigQuery features are covered by the certifications and security statements of Gemini for Google Cloud with exception of the following limitations:

To learn more about certifications and security for Gemini for Google Cloud, see certifications and security for Gemini for Google Cloud.

Secure and responsible use

You should adhere to the following best practices to help ensure the secure and responsible use of Gemini in BigQuery:

  • Use IAM to give the least privilege necessary. For information about security best practices in BigQuery, see Introduction to security and access controls in BigQuery.
  • Be mindful of the data you include in your natural language prompts in BigQuery, such as sensitive or personal information.
  • Review and validate the responses generated by Gemini in BigQuery. Always treat AI-generated code and analysis as suggestions that require human review.
  • Only enable Gemini in BigQuery for projects that don't require compliance offerings other than those listed previously and by Gemini for Google Cloud. For information about how to turn off or prevent access to Gemini in BigQuery, see Turn off Gemini for Google Cloud products.

What's next