Claude Data Retention Explained: 30-Day Rules, ZDR and Project Settings

Why “Claude keeps data for 30 days” is too simple

The phrase 30-day retention appears repeatedly in Anthropic’s documentation, but it describes several different mechanisms. One rule applies to ordinary Anthropic API requests. Another describes what happens after a user deletes a saved conversation. Enterprise administrators can set their own retention periods, with 30 days as the minimum. Certain Covered Models can also require 30-day retention even for organizations that otherwise use ZDR.

The practical lesson is to ask three questions before interpreting “30 days”: Which Claude product are you using? What retention configuration is attached to that product or workspace? And are you using a feature or model with its own storage requirement?

The standard Anthropic API rule: deletion within 30 days

For normal Anthropic API use, Anthropic says it automatically deletes inputs and outputs from its backend within 30 days of receipt or generation. This is the baseline commercial API rule, not a promise that every related object in every Anthropic feature disappears in exactly 30 days.

Anthropic lists several exceptions. Data can remain longer when you use a service that intentionally stores it under your control, such as the Files API; when your organization has a different contractual arrangement such as ZDR; when Anthropic needs to retain information longer to enforce its Usage Policy; or when retention is required by law.

Official source: Anthropic Privacy Center: How long do you store my organization’s data?

What Zero Data Retention actually means

Zero Data Retention is not a switch available to every Claude user. Anthropic describes ZDR as an approved commercial arrangement for some Claude Platform and Claude Code customers. Under an eligible ZDR arrangement, Anthropic does not store covered inputs or outputs except where storage is necessary to comply with law or combat misuse or harm. Anthropic still retains User Safety classifier results for Usage Policy enforcement.

ZDR therefore changes the default 30-day API posture for eligible traffic, but it does not mean “Anthropic retains absolutely nothing under any circumstance.” Safety metadata, legal obligations and features that inherently need persistent storage remain important exceptions.

Anthropic says ZDR requests are reviewed and applied per organization. Claude Platform customers can check whether ZDR is active under Settings → Privacy Controls → Data retention period.

Official ZDR scope: I have a zero data retention agreement with Anthropic. What products does it apply to?

What products does Anthropic say ZDR covers?

Anthropic’s current commercial privacy guidance limits ZDR to specific surfaces. The main eligible categories are eligible Anthropic APIs, Anthropic products that use the organization’s commercial API key, including Claude Code accessed through that API route, and Claude Code for Enterprise where the approved arrangement applies.

That means an organization should not assume that a ZDR agreement for its API automatically turns every Claude interface into a zero-retention product. Claude chat, ordinary Claude for Work experiences, Workbench-style persistent surfaces, Files API storage, Batch API workflows and other stateful features can have different rules.

For regulated workloads, the correct question is therefore not simply “Do we have ZDR?” but “Is this exact model, feature, workspace and access path covered by our approved ZDR configuration?”

Covered Models can require 30-day retention even in a ZDR organization

Anthropic introduced a separate retention rule for certain high-capability Covered Models. Its current documentation says prompts and outputs for designated covered models are retained for 30 days to support safety work on platforms where the requirement applies.

The policy is especially important for organizations that previously configured ZDR. Anthropic’s September documentation says Covered Models such as Claude Fable 5.1 can require 30-day retention and are not available from a zero-retention workspace unless Anthropic has expressly authorized the organization, for example under Enterprise Frontier Safeguards.

This is more nuanced than saying “Fable can never run under ZDR.” Anthropic explicitly notes that some organizations may receive authorization that preserves ZDR access under additional safeguards. Organizations without that authorization need to turn on 30-day retention for the relevant API workspace before Covered Model requests will succeed.

When retention is enabled for one non-default API workspace, Anthropic says the change applies to new requests from that workspace and does not automatically change the organization’s other ZDR-enabled workspaces. The Console displays the state as either Zero data retention or 30 day retention.

Current Covered Model guidance: Data retention practices for Covered Models and Turn on data retention for a Workspace in a zero data retention organization.

Claude Enterprise can use custom retention periods

Claude Enterprise adds an administrator-controlled retention layer for saved conversations and projects. Anthropic says a Primary Owner or Owner can configure custom retention periods under Organization settings → Data and Privacy.

The minimum custom period is 30 days. Anthropic counts each configured month as 30 days, so a three-month policy is treated as 90 days. If the organization does not configure custom retention, Anthropic says the relevant Enterprise data is retained indefinitely by default.

Deletion is scheduled for midnight UTC on the relevant day. Anthropic warns that reducing a retention period can immediately schedule older data for permanent deletion, with a background process carrying out deletion. Once data has passed the retention period and been permanently deleted, Anthropic says it cannot be recovered.

Official Enterprise policy: Configure custom data retention controls for Enterprise plans.

How retention works inside Claude Projects

Projects are where the retention model becomes easiest to misunderstand. Anthropic says project retention takes precedence over chat retention for chats stored inside a project.

Consider an Enterprise organization that sets a 30-day retention period for standalone chats but does not set a custom project retention period. Anthropic says projects are retained indefinitely by default, so chats inside those projects are not deleted merely because the standalone-chat period has elapsed.

If the organization does set a project retention period, the project’s clock is based on its last update. Anthropic explicitly includes events such as creating a new chat inside the project or modifying the project knowledge base. Those actions can move the project’s retention date forward.

Moving a chat changes which rule applies. A standalone chat moved into a project becomes subject to that project’s retention period. A chat moved out of a project becomes standalone again and can become eligible for deletion under the standalone-chat policy based on its own last activity.

Project settings and retention settings are different controls

The title “Project Settings” can suggest that every Claude project has its own little retention dropdown. Anthropic’s public documentation does not describe the normal Enterprise retention policy that way. The custom retention period is configured at the organization level, and that policy then determines when projects expire according to project activity.

Individual project controls focus on a different question: who can see and use the project? Team and Enterprise projects can be kept private or shared more broadly, and project owners can grant members view or edit permissions. Those visibility and sharing settings do not replace the organization’s data-retention policy.

Anthropic’s project help page also distinguishes archiving from deletion. Archiving is an organizational tool: it preserves project knowledge, conversations, members and permission levels. An archived project remains accessible and must be unarchived before it can be deleted.

Project management source: How can I create and manage projects?

What happens when you manually delete a Claude conversation or project?

For commercial saved conversations, Anthropic says deleting a chat removes it from the visible chat history immediately and deletes it from backend storage systems within 30 days, subject to safety and legal exceptions.

A project is broader than a single conversation because it can include project knowledge, instructions, chats and artifacts. Anthropic’s Enterprise retention controls state that when a project reaches the end of its configured retention period, the project and the chats and artifacts within it are deleted.

Manual project deletion is available from the project menu. Archived projects first have to be unarchived before the delete option appears. That is an interface rule, not a retention extension: archiving alone should not be interpreted as a privacy or deletion control.

For ordinary consumer accounts, deleting a chat follows a similar visible/backend split: it disappears from history immediately and is automatically deleted from backend systems within 30 days, subject to Anthropic’s stated exceptions.

Data retention is not the same thing as model training

A company can retain data for operational or security reasons without using that data to train a general model. Anthropic’s commercial policy says that, by default, inputs and outputs from commercial products such as Claude for Work and the Anthropic API are not used to train Anthropic’s models.

Commercial data can be treated differently when a customer explicitly submits feedback, reports a bug, or otherwise chooses to allow Anthropic to use the information. Anthropic says feedback can include the related conversation and can be retained for up to five years.

This is why a 30-day retention setting should not be read as “Anthropic trains on my data for 30 days.” Retention answers how long information is kept. Training answers whether that information is used to improve models. They are separate policy questions.

Commercial training policy: Is my data used for model training?

Claude Free, Pro and Max use a different privacy framework

Consumer Claude plans should not be analyzed using the commercial API policy. Anthropic’s consumer privacy documentation says saved chats remain available to the user until they are deleted. When a consumer deletes a conversation, it is removed from chat history immediately and deleted from backend systems within 30 days.

The major consumer-specific difference is Model Improvement. If a consumer allows Anthropic to use chats or coding sessions to improve Claude, Anthropic says it may retain that data in de-identified form in model-training pipelines for up to five years. The setting applies to new or resumed chats after the model-improvement option is enabled.

Anthropic says turning Model Improvement off stops previous and new chats from being used for future training, while deleting a chat also prevents that chat from being used to train future models. Data already included in a training run or a completed model is not retroactively removed from that model. Incognito chats are not used to improve Claude even when Model Improvement is enabled.

Consumer sources: How long do you store my data? and Is my data used for model training?

Safety, feedback and legal rules can extend retention

Nearly every short retention statement in Anthropic’s documentation comes with exceptions. For commercial and consumer products, Anthropic says conversations or sessions flagged by automated trust-and-safety systems as violating its Usage Policy can be retained longer.

Anthropic’s current commercial policy states that flagged inputs and outputs can be retained for up to two years, while trust-and-safety classification scores can be retained for up to seven years. Feedback submissions can be retained for five years. Anthropic also reserves the ability to retain data when required by law and, where its contracts permit, may keep anonymized information longer for research or statistical purposes.

Covered Models add another safety-specific category. Anthropic says 30-day retention for those models is designed to detect harmful patterns across requests. Human review is not the default; Anthropic says retained content can be accessed through a controlled process when automated systems flag potential harm, with access restricted to approved reviewers and logged.

Those exceptions are why security and compliance teams should base policy decisions on the exact contract and Anthropic configuration that applies to their organization, rather than relying on a single marketing phrase such as “30 days” or “zero retention.”

Which Claude retention rule applies to you?

For an Enterprise audit, it is also worth recording who can change retention, the active chat and project periods, which projects are still active, which API workspaces have ZDR or 30-day retention enabled, and whether any teams rely on Covered Models or persistent API features. That produces a much clearer data map than treating Claude as one homogeneous service.