What is zero data retention? How six AI providers define it, and what they still keep

The Morse teamPublished October 19, 20267 min read

Zero data retention (ZDR) is a promise from an AI provider that the prompts you send and the replies its model returns are not stored after your request is processed. As of October 2026, OpenAI, Anthropic, Google Cloud, xAI, Together AI, and OpenRouter all offer some form of it to developers and businesses, but each sets its own terms for who qualifies, and each keeps something back.

We read each provider's own documentation on October 19, 2026 and set the results side by side below. ZDR is a real protection: what was never stored can't leak in a breach, be used for training, or be handed over later. It's also narrower than the name suggests, and it lives in API settings and contracts, not in the chat apps most people use.

What does zero data retention mean in practice?

It means the content of your request isn't written to storage. The provider still has to read your prompt to answer it. Together AI's ZDR page describes it this way: request content "may pass through volatile memory and temporary caches while your request is served, but it is not persisted to non-volatile storage." OpenRouter's definition goes further: a provider "will not store your data for any period of time." Even so, OpenRouter doesn't count in-memory prompt caching as retention.

Three things usually fall outside ZDR:

• Metadata. Together AI still keeps "token counts, model IDs, and timestamps" for billing, along with your account and billing details (Together AI).
• Things you store on purpose, like uploaded files, saved conversations, and batch jobs.
• Safety exceptions each provider writes for itself, covered below.

Which AI providers offer zero data retention, and who can get it?

Each provider grants ZDR its own way. The table sets out what each one's own pages said on October 19, 2026.

OpenAI says ZDR is "subject to prior approval by OpenAI and acceptance of additional requirements" (OpenAI). With it on, the store setting on the Chat Completions and Responses APIs is always treated as false. OpenAI also reserves the right to make specific models ineligible for ZDR for specific customers, with written notice. In August 2026 it announced Private Safety Processing, which it says lets automated systems flag misuse across related requests "without giving OpenAI personnel access to the underlying content," so it can keep offering ZDR on more capable models.

Anthropic moved the other way for its most capable models. Since June 9, 2026, prompts and outputs on covered models "are retained for 30 days to support our safety work, on every platform where these models are offered," so a ZDR organization has to turn retention on for any workspace that uses them (Anthropic). Anthropic says no staff can read that retained data by default, and that human review happens "only through a controlled access path."

Google treats ZDR as a checklist. Its page says "to achieve zero data retention, customers must take specific actions," including turning off caching, leaving request-response logging off, setting store to false, and asking for an exception to abuse monitoring (Google Cloud, updated October 15, 2026). For Search grounding, "there is no way to disable the storage of this information." Google's separate Gemini API terms say that on paid plans it logs prompts and responses "for a limited period of time" to enforce its policies, and we didn't find a ZDR option in those terms.

xAI warns against it. Its docs say "for most customers, we do not recommend enabling ZDR," because it switches off the Files, Collections, and Batch APIs and server-side conversation history, and because content deleted under ZDR "can never be recovered or produced" (xAI).

Does zero data retention stop a provider from training on your data?

It does, since nothing is kept to train on, but the reverse doesn't hold. Training is usually a separate setting. Together AI says training on your data "is a separate opt-in and is off by default," with or without ZDR (Together AI). OpenAI says API data hasn't been used for training since March 1, 2023 unless you opt in, yet it still keeps abuse logs for up to 30 days by default (OpenAI). OpenRouter warns that some endpoints that don't train "do retain it (e.g. to scan for abuse or for legal reasons)" (OpenRouter).

What doesn't zero data retention cover?

Four gaps come up across the six providers:

• Safety carve-outs. OpenAI keeps images flagged as potential CSAM even under ZDR, and Anthropic's covered models need 30 days of retention.
• Search and tools. Google keeps Search grounding data for up to three days, OpenRouter's ZDR covers only "provider routing for inference requests," and OpenAI says data sent to third-party MCP servers "is subject to their data retention policies."
• Stored features. Saved conversations, uploaded files, vector stores, and batch jobs keep data by design. OpenAI excludes them from ZDR, Together AI keeps uploaded files until you delete them, and xAI switches those features off.
• Consumer apps. ZDR is a setting on an API account. Typing into ChatGPT, Claude, or Gemini runs under each app's consumer retention rules instead, which we compare in are AI chats private.

Can you check that a provider keeps nothing?

Mostly, you can't. ZDR is a written commitment, and the provider's systems are out of your sight. Of the six providers' pages we read, only xAI's describe a signal on every reply: each API response carries an x-zero-data-retention header set to "true" or "false," so an app can confirm the setting is on (xAI). The header shows the setting is on, but it can't show what happens on xAI's servers. OpenRouter makes the same point about retention declarations on private endpoints: they are "not independent verification" (OpenRouter).

When we searched this question on October 19, 2026, the AI answer at the top of Google's results said prompts and responses under ZDR are "deleted within seconds after the response is returned." The providers' own pages list longer exceptions: three days for Google's Search grounding, 30 days on Anthropic's covered models, and flagged images at OpenAI.

How Morse uses zero data retention

Zero data retention is a contract between a provider and a business. OpenAI grants it by approval and Anthropic by agreement, and none of it reaches the chat apps people type into. Morse has zero data retention agreements with its inference, search, and other providers, so Private-mode chats get ZDR without you negotiating it, and your chat history never sits on a server to begin with.

• Private, for zero data retention by default. Private is the default mode for a new chat. Open-weight models like DeepSeek, Kimi, and GLM, plus xAI's Grok, run under zero data retention agreements Morse has with its inference, search, and other providers. Where a provider confirms zero retention on each reply, Morse checks it and refuses a reply that doesn't. Otherwise, Morse relies on the provider's commitment.
• E2EE, for when a promise isn't enough. ZDR still lets the host read your prompt while it answers. In E2EE, your browser checks a secure enclave's hardware attestation, then encrypts the message so only that verified enclave can open it, and Morse's servers relay ciphertext they can't read. The enclaves are run by third-party providers, and the guarantee rests on the enclave hardware and attestation working as designed. E2EE chats, with models like GLM 5.3 Flash, come with Pro, at $19 a month.
• Anonymous, for GPT, Claude, and Gemini. Morse's proxy removes your account, email, and IP address and sends the message under Morse's own account, so the provider answers without knowing it's you. These providers may keep the message for a limited time to check for abuse, under their own policies, so for a prompt you'd want kept nowhere, choose Private or E2EE.

In every mode, Morse's servers don't keep your messages, and your chat history is stored only in your browser, encrypted. Morse doesn't train AI on your chats, sell your data, or show ads (privacy policy). To see how this compares with other ways of keeping AI chats private, read our guide to using AI privately, or see how DeepSeek's open weights avoid DeepSeek's own servers.

Start free, and your next chat opens in Private, with zero data retention, by default.