What Claude Code zero data retention means
Claude Code zero data retention is a setting on a Claude for Enterprise organization that stops the prompts you send and the responses Claude generates in Claude Code sessions from being stored once the response is returned. The model calls are processed in real time and that is it. The one exception: data can still be kept where the law requires it or to deal with a usage policy violation. If a session is flagged for a violation, its inputs and outputs can be kept for up to 2 years.
For comparison, the standard for commercial accounts (Team, Enterprise and API) is a 30-day retention period, and commercial traffic is not used to train models unless the organization opts in to a program that provides data for that. So the question ZDR answers is not "will my code train a model". On commercial terms it already does not. The question is whether anything sits on the provider's servers at all after a request finishes.
Who can get ZDR and how to request it
ZDR is not part of the standard Claude for Enterprise plan, and there is no switch for it in your admin settings. It is offered to qualified accounts. You ask sales or your account team, they submit the request, and it gets enabled on your organization once eligibility is confirmed. Every enablement action is audit-logged.
- It is per organization. If you create a second organization under the same account, it does not get ZDR automatically. Ask for it again.
- It covers the direct platform only. If you run Claude Code through Amazon Bedrock, Google Cloud's Agent Platform or Microsoft Foundry, that provider's own retention policy applies instead. See Claude Code on Bedrock, Vertex AI and Microsoft Foundry for those setups.
- If you already use ZDR through pay-as-you-go API keys, you can move to Claude for Enterprise and keep ZDR for Claude Code, which also gets you the admin features: per-user cost controls, the analytics dashboard, server-managed settings and audit logs.
If your reason for wanting ZDR is health data, check the HIPAA route too. An Enterprise organization with HIPAA enabled can bring the Claude Code CLI and the Code tab in the desktop app under its business associate agreement without ZDR, once the HIPAA configuration is applied to local mode. Without that configuration, Claude Code still needs ZDR for that coverage.
What ZDR covers and what it does not
ZDR covers model inference made through Claude Code on Claude for Enterprise, for every model your ZDR organization can use. That is the core of what your developers do all day: the prompts, the file contents Claude reads into context and the code it writes back. It does not reach everything around that.
Not covered by Claude Code ZDR
| Area | What happens instead |
|---|---|
| Chat on claude.ai | Conversations in the Enterprise web interface follow standard retention |
| Cowork | Cowork sessions follow standard retention |
| Claude Code analytics | No prompts or responses are stored, but productivity metadata such as account emails and usage statistics is collected. Contribution metrics are not available, only usage metrics |
| User and seat management | Account emails and seat assignments are kept under standard policies |
| MCP servers and other integrations | Whatever a third-party tool or MCP server does with data is outside ZDR. Review each one yourself |
The last row is the one people miss. ZDR stops the model provider from keeping your prompts, but if Claude Code sends a customer record to a ticketing MCP server, that server's retention rules decide what happens to it. This is exactly why I would pair ZDR with an approved MCP list. Claude Code managed MCP covers how to build one.
Features ZDR turns off
Anything that only works by storing prompts or completions is disabled for a ZDR organization. This happens on the backend, so it holds no matter what the client shows. If someone tries one of these, they get an error saying the organization's policies do not allow that action.
Disabled under Claude Code ZDR
| Feature | Why it is off |
|---|---|
| Cloud sessions, including ones started from the desktop app | Session history with prompts and completions is stored server-side |
| Chat and Cowork unified | It depends on cloud sessions |
| Claude Tag | It keeps channel memory and session transcripts |
| Artifacts | Published page content is stored on hosted infrastructure |
| /feedback, /bug and /share | Submitting feedback sends the conversation to the provider |
| Remote Control | The transcript is stored on servers to sync across devices |
Tell your team before ZDR lands. Local sessions keep working, in the desktop app or the terminal, but anyone who relies on Claude Code on the web or Remote Control to check in from their phone will lose it. Expect new features to land disabled too if they need stored transcripts.
One model note. The Fable models require data retention by default, so whether a ZDR organization can use them is decided by that model policy, not by Claude Code. Where they are not allowed, they are missing or greyed out in /model and the server rejects requests for them. The best alias then resolves to Opus instead.
Make sure your team's traffic actually lands in ZDR
ZDR applies to requests that authenticate into the ZDR-enabled organization. A developer who signs in with a personal account, or uses an API key from another organization, is not covered, and nothing in the session will tell them. This is the gap I would close first.
- Find your organization ID in claude.ai admin settings.
- Set forceLoginMethod to "claudeai" and forceLoginOrgUUID to that ID in managed settings, and push them with your device management tooling (Jamf, Intune, Group Policy or your Linux equivalent). Server-managed settings only arrive after someone has signed in to your organization, so they cannot steer a first login.
- If you also deliver server-managed settings from claude.ai, set both keys there too until every machine runs Claude Code v2.1.299 or later.
- Check the side doors. ANTHROPIC_API_KEY, ANTHROPIC_AUTH_TOKEN and apiKeyHelper credentials are blocked at startup under forceLoginOrgUUID, which is what you want. But claude setup-token and /install-github-app only enforce the login method, so they can mint a token in a different organization. Decide who is allowed to run those.
- Test on one managed machine: sign in with a personal account and confirm Claude Code refuses it and exits.
My rollout order for a ZDR team
- Confirm ZDR is enabled on the exact organization your developers will sign in to, and on any new one you create later.
- Deploy managed settings with the login lock and a cleanupPeriodDays value that matches your policy.
- Lock down MCP to an approved list, since third-party servers sit outside ZDR.
- Announce what goes away: cloud sessions, Remote Control, Claude Tag, artifacts and /feedback.
- Use OpenTelemetry for the deeper numbers, because contribution metrics are not available in the analytics dashboard under ZDR. Claude Code OpenTelemetry walks through the setup, and you control where that data goes.
My take: ZDR is worth asking for if your contracts or your security team need a no-server-side-storage answer, and it is overkill if standard commercial retention already satisfies them, because you give up cloud sessions and Remote Control for it. Either way, the login lock is the step that decides whether ZDR protects anything. If you are still choosing how to deploy, running Claude Code through an LLM gateway is the other big architecture choice worth reading.
Short, practical drops on skills, MCP, agents, prompts, and more. No spam, unsubscribe anytime.
Frequently asked questions
Does Claude Code have zero data retention?
Yes, for qualified organizations on Claude for Enterprise. With ZDR enabled, prompts and responses from Claude Code sessions are not stored after the response is returned, except where required by law or to address usage policy violations.
How do I turn on zero data retention for Claude Code?
You cannot turn it on from admin settings. Contact sales or your account team, who submit the request and enable it on your organization after confirming eligibility. Each organization has to be enabled separately.
What features stop working under ZDR?
Cloud sessions, including ones started from the desktop app, Chat and Cowork unified, Claude Tag, artifacts, /feedback, /bug and /share, and Remote Control. They are blocked on the backend and return a policy error.
Does ZDR cover MCP servers and claude.ai chat?
No. ZDR covers Claude Code inference. claude.ai chat, Cowork, seat and account data, analytics metadata and anything processed by MCP servers or other third-party integrations follow their own retention rules.
Is Claude Code ZDR available on Bedrock or Vertex AI?
Claude Code ZDR on Claude for Enterprise applies to the direct platform only. On Amazon Bedrock, Google Cloud's Agent Platform or Microsoft Foundry, that provider's data retention policy applies.
Last reviewed by David Iya on November 7, 2026
