Last Updated: May 6, 2024
Welcome to Snowflake’s AI Trust and Safety page. Here we address your questions regarding the use of “Snowflake AI Features,” as described in the Documentation, including Snowflake Cortex LLM Functions, Universal Search, and Snowflake Copilot. All capitalized terms used but not defined on this page have the meanings ascribed to such terms in the software-as-a-service or cloud agreement, Terms of Service, or other agreement covering your use of the Service (the “Customer Agreement”). For clarity, any Snowflake AI Features that are made available as preview features are subject to the Preview Terms.
We host and/or manage three types of artificial intelligence models that power the Snowflake AI Features: our own proprietary models, open-source models, and licensed proprietary models (collectively, “Models”). Snowflake AI Features are subject to our Acceptable Use Policy and Shared Responsibility Model for data protection, governance, and security. We understand that trust is the foundation of our customer relationships and are committed to maintaining high standards of data security and privacy.
Is Customer Data used as training data for the LLMs?
No, we do not use your Customer Data to train any Model to be made available for use across our customer base and do not use your Customer Data to fine-tune any Model except at your election or direction. Fine-tuned Models built using your Customer Data can only be used by you.
How are customer inputs and outputs treated when using the Snowflake AI Features?
The treatment of inputs and outputs of Snowflake AI Features depends on whether they are classified as “Customer Data” or “Usage Data” under your Customer Agreement. The classification of inputs and outputs are set forth on the “Snowflake AI Features” section of the Documentation.
Does Snowflake indemnify customers for copyright infringement claims?
Yes, if you are a party to our AI Terms (available under Offering-Specific Terms), in addition to the Snowflake indemnification obligations set forth in your Customer Agreement, Snowflake’s obligation to defend and indemnify you against third-party intellectual property claims (including copyright infringement claims) also applies to outputs from a generally available Snowflake AI Feature. If you are unsure whether you are a party to our AI Terms, please contact [email protected].
Does my Customer Data leave the Service boundary when using Snowflake AI Features?
No.
How does Snowflake ensure that my Customer Data remains secure when using Snowflake AI Features?
Snowflake AI Features are part of the Service and therefore process Customer Data in accordance with our standard security program in place for the Service. We implement robust data protection features like encryption designed to protect the confidentiality, integrity, availability, and security of your Customer Data. Under Snowflake’s Shared Responsibility Model, you are responsible for managing access to your Customer Data. Snowflake supports multiple security methods, including MFA, SSO, Federation and others.
Does Snowflake allow customers to create fine-tuned Models? If so, are these usable or accessible across all Snowflake customers?
We may enable you to fine-tune Models in connection with certain Snowflake AI Features. If you elect to fine-tune a Model, such fine-tuned Model may only be used by you and will not be available to other Snowflake customers.
If I bring a model into the Service (e.g., in Snowpark Container Services) apart from the Snowflake AI Features, how does Snowflake treat it?
Any model (including third-party models) that you bring into your Service account (for example, using Snowpark Container Services) in accordance with the Documentation are treated as Customer Data and are not considered part of the Service. As Customer Data, these models will be subject to the rights and obligations applicable to Customer Data, as set forth in your Customer Agreement.
What are the sources of the Models leveraged by Snowflake in the Snowflake AI Features?
As noted above, Models may include our own proprietary models, open-source models, and proprietary models licensed from third parties. In some instances, we have fine-tuned open-source models for your use as part of the Snowflake AI Features. In other instances, Snowflake is providing the open-source models "as-is" for your use. Please see the Documentation for more details.
Can customers decide not to use Snowflake AI Features?
Yes, Snowflake AI Features are optional for Snowflake customers to use.
Can customers disable Snowflake AI Features at the account level?
The ability to disable Snowflake AI Features varies depending on the feature. For detailed queries regarding particular Snowflake AI Features, please review the Documentation or contact our support team for guidance.
Is there a risk of a Model “remembering” something from another Snowflake customer’s Customer Data?
No, when using Snowflake AI Features, Customer Data inputs are isolated at the software level, and there is no commingling of Customer Data across Snowflake customers.
Who can customers contact for any further questions or concerns about AI trust and safety at Snowflake?
If you have any further questions or feedback, please do not hesitate to contact your account team, or open a support ticket. Please see this link for instructions on opening a support ticket.