AI API data terms: training, retention and zero retention
What does each AI provider say happens to data sent to its commercial API: is it used for training, how long is it kept, and can zero retention be had?
Commercial API offerings from major AI model providers and hosts, with a separate row where a provider has routes with distinct terms, such as free and paid tiers. Records what each provider's own pages state, not what happens in practice. Consumer apps are left out. A blank means we couldn't find it in the provider's own pages, not that the answer is no. A cell holds one passage, so a provider may have more exceptions than the one shown: open the source. Each row reflects the page as read on the date shown. Not legal advice.
Last read 2026-11-01. "Not found" means no source we could use said so, which never means no. "None stated" means the source was read and gives no such value. Each row's page gives the source and exact quote for its values.
| Name | Provider | Offering | Trains on API data by default | Default retention | Exceptions | Zero retention offered | How to get it | BAA available | Data location |
|---|---|---|---|---|---|---|---|---|---|
| OpenAI API | OpenAI | OpenAI API | No | Abuse monitoring logs are generated for all API feature usage and retained for up to 30 days | We reserve the right to make models ineligible for zero data retention or modified abuse monitoring for specific customers | Yes | Prior approval by OpenAI and acceptance of additional requirements | Yes | Data residency controls are a project configuration option; contact our sales team to see if you're eligible |
| Anthropic Claude API | Anthropic | Claude API | No | Automatically delete inputs and outputs on our backend within 30 days of receipt or generation | Claude Fable 5.1, Claude Mythos 5.1, Claude Fable 5, and Claude Mythos 5: these models require 30-day data retention and are not available under ZDR unless expressly authorized by Anthropic | Yes | Reach out to our sales team; requests are reviewed and applied on a per-organization basis | Yes | Inference geo: only "us" and "global" are available. Workspace geo: only "us" is currently available |
| Google Gemini API (paid services) | Paid quota of the Gemini API | No | Fifty-five (55) days for the purposes of detecting and preventing violations of the prohibited use policy | Grounding with Google Search: stores prompts, contextual information, and generated output for thirty (30) days; no way to disable | Not found | If your workload requires guaranteed zero data retention or enterprise data processing agreements, use Vertex AI | Not found | May be stored transiently or cached in any country in which Google or its agents maintain facilities | |
| Google Gemini API (unpaid services) | Unpaid quota in Gemini API | Yes | Fifty-five (55) days for the purposes of detecting and preventing violations of the prohibited use policy | In the European Economic Area, Switzerland, or the United Kingdom, the paid services terms apply to all services, including unpaid quota in the Gemini API | Not found | Not found | Not found | Not found | |
| Google Vertex AI (Gemini Enterprise Agent Platform) | Gemini Enterprise Agent Platform | No | Customer data is retained for limited periods of time in the following scenarios and conditions | Models designated as "advanced AI": all prompts and responses will be logged and securely stored for up to 30 days | Yes | Customers must take specific actions within each of these areas | Yes | Data stored at rest in the customer selected location remains at rest in that location. Requests submitted to a global endpoint may be processed in any Google Cloud location around the world | |
| Azure OpenAI / Foundry Models sold by Azure | Microsoft Azure | Foundry Models sold by Azure | No | Not found | Azure preview features, including models sold by Azure in preview, may employ different privacy practices, including with respect to abuse monitoring | Not found | Customers who meet additional limited access eligibility criteria apply to modify abuse monitoring by completing this form; some advanced models may have more stringent criteria | Not found | Processed within the customer-specified geography (unless you are using a global or datazone deployment type) |
| Amazon Bedrock | Amazon | Amazon Bedrock | No | Actual retention depends on the model - consult the model's terms for specifics | For Anthropic Claude Fable 5 and Claude Fable 5.1, all traffic will be retained for up to 30 days for automated offline abuse detection | Yes | Set data_retention_mode to none | Not found | Encrypted and stored at rest in the AWS Region where you are using Amazon Bedrock |
| Mistral Studio (free mode) | Mistral | Mistral Studio free mode | Yes | For the period necessary to generate the output and then for thirty (30) rolling days to monitor abuse | If you use our Agents API, we keep your input and output until you terminate your account | No | ZDR is available only with pay-as-you-go. If you're in free mode, enable pay-as-you-go first | Not found | By default, your data is hosted in the European Union. You may explicitely use our US API endpoint and your data is hosted in the United States |
| Mistral Studio (pay-as-you-go) | Mistral | Mistral Studio pay-as-you-go | Not found | For the period necessary to generate the output and then for thirty (30) rolling days to monitor abuse | The opt-out preferences you selected (including through zero data retention) does not apply to labs or preview models | Yes | Provide sufficient detail of your legitimate reasons; we will review your request and, at our discretion, approve or deny it | Not found | By default, your data is hosted in the European Union. You may explicitely use our US API endpoint and your data is hosted in the United States |
| Cohere SaaS platform | Cohere | Cohere SaaS platform | Not found | We automatically delete logged prompts and generations after 30 days | Data you allow us to use for training purposes may be retained separately | Yes | We only allow ZDR for enterprise customers who can make additional commitments about their usage | Not found | Not found |
| xAI API | xAI | xAI API | No | All API requests and responses are stored on our servers (encrypted at rest) for 30 days for auditing purposes in the event of suspected abuse or misuse | ZDR disables features that depend on stored data, including the stateful Responses API, files and collections, and the batch API | Yes | Self-serve: where available, a team admin can turn it on or off directly from the xAI Console | Not found | The default endpoint does not guarantee a processing region. To keep retained request data in the United States, use the US regional endpoint |
| DeepSeek Open Platform | DeepSeek | DeepSeek Open Platform | Not found | We keep this personal data for as long as you have an account. This personal data includes your account personal data, input and payment personal data | If you violate any of our terms, policies or guidelines, we may keep your personal data as necessary to process the violation | Not found | Not found | Not found | We directly collect, process and store your personal data in People's Republic of China |
| GroqCloud | Groq | GroqCloud | No | By default, Groq does not retain customer data for inference requests | We may temporarily log inputs and outputs when troubleshooting errors or investigating suspected abuse; these logs are retained for up to 30 days | Yes | All customers may enable zero data retention (ZDR) in Data Controls settings | Yes | All customer data is retained in Google Cloud Platform (GCP) buckets located in the United States |
| Together AI inference | Together AI | Together's inference services | No | Together stores the prompts you send and the responses models return, and may use them for product improvements | Passthrough models: data sent to them is handled under that provider's own policy | Yes | Open the Privacy section of organization settings. Set Store prompts and model responses to No | Not found | Geo-specific inference keeps inference processing in the United States for select serverless models. It covers only where those requests are processed, not where data is stored, logged, or routed |
| Fireworks AI serverless (open models) | Fireworks | Open models | No | Fireworks does not log or store prompt or generation data for any open models, without explicit user opt-in | Response API retention of up to 30 days when store=true is enabled | Yes | Enterprise account admins can additionally turn on the zero data retention policy to reject requests and jobs that would persist or log customer content | Not found | Pin your account's inference to a region, starting with US-only endpoints |
| Perplexity API | Perplexity | Perplexity API platform | No | Retention, use, and security practices vary by API, model, and account configuration | Some providers require limited retention as a condition of offering their models | Yes | Available only to eligible customers through an account-level configuration under a written agreement with Perplexity | Yes | Not found |