> For the complete documentation index, see [llms.txt](https://docs.soda.io/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.soda.io/soda-documentation/soda-ai/ai-security-compliance-and-privacy.md). # AI security, compliance & privacy Learn what data is sent to a model provider by Soda AI features, what Soda stores, and what your organization remains responsible for. Each Soda AI feature sends requests to a large language model (LLM) hosted by a **third-party provider**, currently **OpenAI**. The model's policies will depend on how you connect to it (via Soda-managed AI, or BYOK). This page covers: * what the requests contain * what Soda and the provider store * which settings admins control * what your organization remains responsible for when it relies on AI output

AI features covered in this page

Soda AI chatContract AutopilotContract Copilot

{% hint style="info" %} [Soda MCP](/soda-documentation/soda-ai/soda-mcp.md) and the [root cause analysis (RCA) agent](/soda-documentation/soda-ai/rca-agent.md) run inside AI tools **you choose and configure**, such as Claude Code. Soda returns data to your agent, and **your agent's provider and your agreement with that provider govern what happens to it next**. {% endhint %} > Learn more about [Soda AI](/soda-documentation/soda-ai.md) and its features. *** ## Model providers Every AI feature runs through a provider. Admins manage providers in the **AI** tab of **Organization Settings**.

Which account receives the requests?

Bring your own key (BYOK)

With bring your own key (BYOK), requests go to your organization’s own OpenAI account, under your agreement with OpenAI.

Enterprise plans use BYOK only.

  • The retention, privacy, and regional settings of your own OpenAI account apply.
  • Soda stores a reference to where the API key lives and never the key itself.

Soda-managed AI

If BYOK is not enabled, requests go by default to an OpenAI account that Soda operates, under Soda’s agreement with OpenAI.
  • Usage is included in your Soda license.
  • Soda chooses the models.

Soda-managed AI provider vs Bring your own key (BYOK)

> Using your own supported provider account, agreement, and data-handling settings, Soda AI can fit easily within **your organization's existing security and governance** arrangements. > > Set up your own provider by [bringing your own key (BYOK)](/soda-documentation/soda-ai/bring-your-own-key-byok.md). ### Data sent to the model provider **Row-level data**, such as individual records or personal data in your datasets, **is not sent** unless an admin enables it for Contract Autopilot. | Feature | Data sent to the model provider | | --------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | [**Soda AI chat**](/soda-documentation/soda-ai.md#chat-interface) | | | [**Contract Copilot**](/soda-documentation/soda-ai/contract-copilot.md) | | | [**Contract Autopilot**](/soda-documentation/soda-ai/contract-autopilot.md) | | * A contract you generate from the chat runs through Contract Autopilot, so the Contract Autopilot row applies to that request. * Contract Autopilot computes its profiling metrics from a sample of about 10,000 rows. **Only the resulting metrics go to the model provider**, not the sample. * Failed rows stored in Diagnostics Warehouse stay in your own data warehouse and **are not available to any AI feature**. > Learn more about [Diagnostics Warehouse](/soda-documentation/diagnostics-warehouse.md). ## Storage and retention **Soda does not store your prompts or chat conversations on its servers**. The chat keeps conversation history in **your browser's local storage**, on your device. The history does not carry over to another browser or device, clearing the browser's site data deletes it, and anyone using the same browser profile can read it. The model provider stores **request data**. The chat and Contract Copilot continue a conversation by referring back to responses stored with the provider, so requests and responses stay in the provider account for that provider's standard retention period. With the Soda-managed provider, that account is Soda's OpenAI project, covered by [OpenAI's API data policies](https://platform.openai.com/docs/guides/your-data). With BYOK, it is your own OpenAI project, under your settings. {% hint style="warning" %} **Soda AI features do not support OpenAI projects with zero data retention**. Zero data retention disables the stored responses that the chat and Contract Copilot depend on. {% endhint %} ## Model training **Soda does not use your data, metadata, prompts, or results to train or fine-tune any AI model**. OpenAI does not train its models on API request data, and Soda's OpenAI project has input and output sharing with OpenAI turned off. Soda improves its AI features through development work that does not use your prompts. In some AI features, you can rate a response, and Soda records the rating in its product analytics. ## Access and human oversight **AI features work only inside an authenticated Soda Cloud session and act with your permissions**. They cannot do anything in Soda Cloud that you could not do yourself. Contract Autopilot publishes each generated contract as soon as its scan succeeds, with no separate approval step. Generated contracts carry a purple checkmark until someone edits them. Review them before you rely on them, especially when a contract gates a pipeline. ## Admin controls Admins with the **Manage Organization Settings** global role control AI features from **Organization Settings**: * In the **Organization** tab, under **Soda AI**, enable or disable each AI feature. * For Contract Autopilot, enable or disable **Send row-level data to Soda**, which controls the 100-row sample. * In the **AI** tab, add your own OpenAI provider and choose the provider for each capability. > Contact to request access to an AI feature or setting that does not appear in your organization. ## Terms and compliance Your agreement with Soda governs your use of Soda AI features, and nothing on this page changes that agreement. Whether AI features are available to your organization can depend on the AI terms in that agreement. AI processing within the Soda's services detailed on this page runs under the same security controls as the rest of Soda Cloud and is covered by **Soda's SOC 2 Type 2 compliance**. Data is encrypted in transit and at rest. > Read the [Soda Privacy Policy](https://www.soda.io/privacy-policy). {% hint style="info" %} **Model providers operate under their own separate security policies**. Please note **users are responsible for their arrangements with the model provider**, including the provider agreement, the data processing agreement, appropriate account settings, and provider charges. For more information, go to [OpenAI's security information](https://openai.com/security-and-privacy/). {% endhint %} ## Risks and your responsibilities **Language models are probabilistic**, so output varies between requests and can contain errors or omissions. A generated check can look plausible and still be wrong for your data. **Your organization remains responsible for checking that results suit its data and intended use**, including results used in automated workflows. * **Keep meaningful human oversight** of any decision with legal or similarly significant effects. * **Do not enter personal data**, **credentials**, or other **confidential information** in a prompt. Your messages go to the model provider. * Soda monitors its AI features for **availability** and **technical errors**. It does not check whether each individual answer or suggestion is correct.
*** {% hint style="info" %} You are **not logged in to Soda** and are viewing the default public documentation. Learn more about [Licensing & documentation access](/soda-documentation/reference/documentation-access-and-licensing.md). {% endhint %} --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation by asking a question. Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters: ``` GET https://docs.soda.io/soda-documentation/soda-ai/ai-security-compliance-and-privacy.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.