Zero retention after approval, on eligible endpoints
- Who it's for
- Companies building on OpenAI's API that OpenAI approves for zero data retention.
- What's retained
- Without zero data retention, OpenAI keeps abuse monitoring logs for all API use "for up to 30 days". With it, eligible endpoints such as /v1/chat/completions, /v1/responses and /v1/embeddings are covered. Assistants, threads, files, vector stores and batches keep data "Until deleted". An image flagged as possible child sexual abuse material is kept for manual review even with zero data retention.
- How you get ZDR
- Contact OpenAI's sales team. OpenAI says these controls "are subject to prior approval by OpenAI and acceptance of additional requirements." Once approved, a Data Retention tab appears under Settings, Organization, Data controls. OpenAI may make specific models ineligible for a customer, with written notice.
- Source
- Date checked
- 2026-11-10
Zero retention after approval, on eligible features
- Who it's for
- Organizations building on Anthropic's API, including Claude Code used with commercial API keys.
- What's retained
- By default, Anthropic deletes API inputs and outputs "within 30 days of receipt or generation". Under zero data retention, "Anthropic does not store customer prompts or responses at rest after the API response is returned." It covers the Messages and Token Counting APIs for eligible features. It does not cover code execution, Managed Agents or Claude for Excel, or the models Anthropic calls Covered Models unless expressly authorized.
- How you get ZDR
- Contact Anthropic's sales team. "ZDR is enabled per organization." The Claude Team and Claude Enterprise chat apps are not eligible.
- Source
- Date checked
- 2026-11-10
Human-review storage switched off after approval
- Who it's for
- Customers and partners "managed by a Microsoft account team or under an eligible program".
- What's retained
- By default, when content is flagged, "a sample of customer's prompts and completions may be selected for review" and kept in an abuse monitoring data store. With approval for modified abuse monitoring, "the data storage and human review process described above is not performed. However, automated review may still be conducted." Stateful features, such as the Responses API, Assistants threads and stored completions, store data in your own Azure tenant.
- How you get ZDR
- Apply for modified abuse monitoring with Microsoft's form. After approval, check that "ContentLogging" shows "false".
- Source
- Date checked
- 2026-11-10
Zero retention once you switch several features off
- Who it's for
- Google Cloud customers building with Google's models, or with OpenAI models on Google's platform.
- What's retained
- Under zero data retention, Google says it processes requests "without writing prompts, file inputs, or responses to persistent disk." It still uses an in-memory cache that is cleared within 24 hours, and it records operational and billing metadata such as token counts. Google says prompt logging for abuse monitoring may apply under its Cloud terms. Grounding with Google Search keeps logs for up to three days, and Google says there is no way to disable that. Grounding with Google Maps keeps them for 30 days. Deep Research keeps them for seven days, which cannot be disabled.
- How you get ZDR
- For Google's models, switch off caching and request-response logging yourself, set store=false on the Interactions API, and don't use Grounding with Google Search or Maps. For OpenAI models on Google, zero data retention is "available ... on an approved project basis, but is disabled by default."
- Source
- Date checked
- 2026-11-10
Zero retention by account setting, with exceptions for some models
- Who it's for
- Any AWS account. It is a developer platform, so someone has to build or host the app your team uses.
- What's retained
- With the setting at none, "No request or response data is written to durable storage by AWS or shared with the model provider." AWS names some models whose traffic is kept for up to 30 days for automated abuse detection; for OpenAI models other than the gpt-oss family, flagged traffic is kept for up to 30 days. AWS may store and review flagged content only to check for child sexual abuse material. Model providers "don't have access to Amazon Bedrock logs or to customer prompts and completions."
- How you get ZDR
- Self-serve. Set data_retention_mode to none for your account or project, in each Region; new accounts start at inherit. If a model requires retention, "Amazon Bedrock will block the request and return an error." For OpenAI models, "Eligible customers may request full ZDR through their AWS account team." AWS also says customers eligible for its "Enterprise Frontier Safeguards program will receive ZDR through June 30, 2027" on the models whose traffic it keeps for 30 days.
- Source
- Date checked
- 2026-11-10