Generative AI in Google Workspace Privacy Hub

Yes. Gemini for Workspace accesses customer data in order to provide personalized responses, such as summarizing a Google Doc or analyzing data in a Google Sheet. Gemini for Workspace abides by your organization’s existing controls and data handling practices. 

Data stored by the customer through use of Google Workspace services is considered Customer Data as defined by the Cloud Data Processing Addendum (CDPA) and, as such, is governed by the data processing terms set out in the CDPA. Google only accesses and processes Customer Data in order to provide, secure and monitor the relevant Cloud Services and any related technical support as described in Section 5.2 of the CDPA's General Terms.

You can find more information about how we process user prompts in The life of a prompt: Demystifying Gemini. Below is a depiction of the workflow of a user prompt.

Gemini for Workspace does not save prompts or responses. The prompts that a user enters when interacting with Gemini for Workspace are not used beyond the context of the user session. The data disappears after your Gemini session ends, as we explain in The life of a prompt: Demystifying Gemini.

When Google Workspace business, education and public sector customers use Gemini for Workspace, they get the same robust data protection and security standards that come with all Google Workspace services, and the following specific protections:

• Your interactions with Gemini for Workspace stay within your organization. Gemini for Workspace stores any prompts or generated content alongside your Workspace content and does not share them outside your organization. 
• Your existing Google Workspace protections are automatically applied. Gemini for Workspace brings the same enterprise-grade security as the rest of Google Workspace, automatically applying your organization’s existing controls and data handling practices, such as data-regions storage policies and Data Loss Prevention (DLP). 
• Your content is not used for any other customers. None of your content is used for model training outside of your domain without permission.

You can learn more here: 

• How we're protecting your Google Workspace data in the era of generative AI
• Boosting productivity and creativity with Duet AI in Google Workspace while keeping your organization’s data private

The strict data access control model for Gemini for Workspace does not allow for inputs or session content to leak across user boundaries. The model is structured to only present data from a session to that individual user. Thus, confidential information is not only kept private within an organization, it is also kept private between users within an organization subject to the built-in Workspace access control for shared files. Client-side encryption will restrict Gemini for Workspace’s access to sensitive data and can be leveraged to accomplish the same goal.

The prompts that a user enters when interacting with features available in Gemini for Workspace are not used beyond the context of the user session. Prompt content is not used for learning or fine-tuning Google’s AI models. User sessions are considered to be within the Workspace trust boundary and its contents are not shared outside of that specific user session.

The content entered in Gemini for Workspace prompts stays within your organization. Additionally, for supported Workspace editions, existing DLP control will continue to apply to generated output from Gemini for Workspace that is inserted into Workspace services covered by DLP (for example emails, shared Drive files, and Docs). Ultimately no content is shared unless a user in your organization explicitly does so.

Gemini for Workspace only retrieves relevant content in Workspace that the user has access to in order to contextualize the prompt and ground responses. If the user doesn’t have access to a document or email, Gemini for Workspace will not retrieve that content. Content restrictions, such as Information Rights Management (disable download, copy, or print) and client-side encryption, can also help restrict Gemini for Workspace’s access to sensitive data. For example, you can leverage built-in AI classification or DLP capabilities in Workspace to identify sensitive data, automatically apply classification labels, and enforce Information Rights Management (IRM) controls based on the classification labels to restrict Gemini for Workspace from accessing sensitive data for the users under the IRM restriction.

No. The prompts that a user enters when interacting with Gemini for Workspace are not used beyond the context of the user session. User prompts are considered Customer Data under the Cloud Data Processing Addendum. Workspace does not use customer data for training models without customer’s prior permission or instruction. This commitment is outlined in the ‘Training Restriction’ sections of the Google Workspace Service Specific Terms.

At Google, we take our contractual commitments to customers very seriously. Customers can count on Google to handle Customer Data in line with the agreement, including our commitments (1) to process Customer Data according to customer instructions under the agreement, and (2) not to use Customer Data to train or fine-tune any of Google’s generative artificial intelligence models supporting the Google Workspace Generative AI Services without our customer’s prior permission or instruction.

But they aren’t just words. To ensure we continually meet these high standards, independent auditors validate our practices against international standards and best practices. We’ve attained the most comprehensive set of safety, privacy and security certifications and attestations for Gemini for Workspace from internationally recognized regulatory and compliance bodies - HIPAA, ISO/IEC 27001, 27701, 27017, 27018, 9001, and 42001, the world's first international standard for Artificial Intelligence Management Systems (AIMS).

Google's foundational language models are trained primarily on publicly available, crawlable data from the internet, drawn from sources like blog posts and public conversation forums. We give publishers control over how their sites are used with Google-Extended, a control that web publishers can use to manage whether their sites help improve Gemini and Vertex AI generative APIs.

When a user enters a prompt in the Gemini for Workspace side panel, it will “search” for relevant content within your organization that you have access to such as specific emails, relevant presentations and docs and generate a response “grounded” in this information. 

In most cases following this, you will have the option to Retry with Google Search . When you use this option, Gemini for Workspace leverages Google Search to improve the quality of its responses by accessing factual information and relevant sources. [Only relevant keywords from the prompt are sent to Google Search as a search query and Gemini for Workspace will generate a response based on the search results]. The data transmitted to Google Search is ephemeral and processed in strict compliance with Workspace's security measures and contractual commitments, and it is not used for product improvement.

Gemini for Workspace uses a collection of models from Google’s rich repository, including Gemini. With these models, Gemini for Workspace can help you write, visualize, organize, and connect more meaningfully. Read more about it on the Workspace Blog.

As with all Google Workspace Core Services, Google Workspace maintains logs of user activity for their authorized Google Workspace administrator(s). The level of detail and retention period depends on the applicable Google Workspace edition and how the administrator has configured settings. Admin audit logs now include Gemini for Workspace activity (see details here) as well as data access by Gemini for Workspace.

Given the global nature of Google Workspace services, Google maintains facilities in all regions (globally) to store and process Customer Data.  We provide information about the locations of Google’s facilities as well as our subprocessors’ facilities.

Customers with qualifying editions can choose to use the Data Regions feature to select a region (such as Europe) to store their covered Customer Data at rest. This feature currently applies to the Google Workspace Core Services and data set out here (which is reflected in the “Data Regions” Section of the Google Workspace Service Specific Terms). 

Please refer to our Cloud Data Processing Addendum, Workspace Services Summary, Service Specific Terms, and Generative AI Prohibited Use Policy.

Google has a long-standing commitment to global privacy compliance, and generative AI services are no different. We’ve attained HIPAA, ISO/IEC 27001, 27701, 27017, 27018, 9001, and 42001 certifications for Gemini for Workspace. With these certifications, Gemini for Workspace has the most comprehensive set of safety, privacy and security certifications internationally recognized by regulatory and compliance bodies.

The achievement of ISO/IEC 42001, the world's first international standard for Artificial Intelligence Management Systems (AIMS), certifies that Gemini for Workspace has been developed, deployed, and maintained responsibly with appropriate ethical considerations, data governance, and transparency. To date, no other generative AI offering for productivity and collaboration has met this level of recognition, showing that Gemini for Workspace is the first in the industry ready to support businesses and public sector organizations while benefiting all users.

Gemini for Workspace can support HIPAA workloads when requested by our healthcare and life sciences customers. The HIPAA Included Functionality and the Google Workspace and Cloud Identity HIPAA implementation guide have been updated to reflect the inclusion of Gemini for Workspace. This makes Gemini for Workspace available to more organizations, and we are working to expand our compliance support coverage for Gemini for Workspace to align with the full set of Workspace certifications.

Gemini for Workspace has been submitted for FedRAMP High authorization.

Google is committed to developing and deploying AI boldly and responsibly with our European partners. For more information on how we are preparing for AI Act compliance, please see our article Navigating the EU AI Act: Google Cloud's proactive approach.

Gemini for Workspace follows the same privacy and security principles as the rest of the Google Workspace suite. You can find the details in our privacy and security support pages.

We published this Generative AI in Google Workspace Privacy Hub to provide our customers, especially those in the public sector, with Gemini for Workspace specific content for their required assessments. Please refer to our DPIA Resource Center for guidance around the DPIA process itself.