What you will be able to do
- State how long deleted chats, API inputs and outputs, and flagged conversations are retained
- Describe who can set Enterprise custom retention, its minimum, and what happens to data when the period ends
- Identify what must be in place before health data or EU personal data goes into Claude, and which features fall outside that cover
1.How long data is kept, and what deletion does
Training is one question. Storage is another, and the two have separate articles for consumer and commercial accounts. A few numbers are the same in both.
When you delete a conversation, on a consumer or a commercial plan, it disappears from your chat history immediately and is deleted from Anthropic's back-end storage within 30 days. For Anthropic API users, inputs and outputs are deleted automatically within 30 days of receipt or generation. The exceptions are a feature with longer retention under your control (such as the Files API), a zero-data-retention agreement, Usage Policy enforcement, or legal requirements.
Two cases override these defaults. First, if automated trust-and-safety systems flag a chat as violating the Usage Policy, inputs and outputs are retained for up to 2 years and classification scores for up to 7 years. This applies on both consumer and commercial plans. Second, certain models, called Covered Models, require 30-day retention of prompts and outputs as part of Anthropic's safety work. On the API, that is the one exception to conversation content not being retained by default.
For a person who has just pasted something they shouldn't have, the practical step is to delete the chat. It leaves the visible history at once and back-end storage within 30 days. On a consumer account it also keeps the chat out of future training.
| Situation | Plan | Retention |
|---|---|---|
| Deleted conversation | Consumer and commercial | Removed from history immediately; back-end within 30 days |
| API inputs and outputs | Commercial (Anthropic API) | Deleted within 30 days unless an exception applies |
| Model improvement enabled | Consumer | De-identified, up to 5 years in training pipelines |
| Flagged as violating the Usage Policy | Consumer and commercial | Inputs/outputs up to 2 years; classification scores up to 7 years |
| Feedback via thumbs up/down | Consumer and commercial | 5 years |
Checkpoint 1 of 4· Check yourself
An API customer has no special agreement. One request is flagged by automated trust-and-safety systems as violating the Usage Policy. How long may its inputs and outputs be kept?
Flagged content is an exception to the 30-day default. Inputs and outputs may be kept for up to 2 years, and the 7-year figure applies only to classification scores.
“We retain inputs and outputs for up to 2 years and trust and safety classification scores for up to 7 years”Source: privacy.claude.com
2.Enterprise custom retention: who sets it and what it deletes
Enterprise organisations can limit how long Claude stores chat and project data. Only a Primary Owner or Owner can set this, so an ordinary member who thinks the period is wrong should go to them rather than look for a personal setting. The minimum period is 30 days, and each month counts as 30 days, so three months means 90 days.
The clock runs from the last activity: the last message in a chat, or the last update to a project. Project retention takes precedence over chat retention, and projects are kept indefinitely by default. That means chats inside a project are not deleted by the chat retention period. Shortening the period schedules any data outside the new window for permanent deletion as soon as the change is saved, and data past its retention period cannot be recovered.
These controls cover content in chats and projects. They do not apply to Claude Design, Claude Tag, Claude Managed Agents, or other features built on Claude Code on the web. Every retention-related change is recorded in audit logs. Separately, a Primary Owner can enable the Compliance API, which lets the organisation pull activity events, chat data and file content into its own monitoring and audit tools. A knowledge worker only needs to know that these oversight tools exist, not how to operate them.
Checkpoint 2 of 4· Put it in order
Put the steps for setting an Enterprise custom retention period in order
- 1.Log in to your Owner Enterprise plan account.
- 2.Set your desired retention period (minimum 30 days).
- 3.Navigate to Organization settings > Data and Privacy.
- 4.Save your changes.
An Owner-level account sets the period under Data and Privacy, and saving it takes effect immediately.
“To set custom retention periods for your organization, you must have either a Primary Owner or Owner role.”Source: privacy.claude.com
3.Regulated data: health information and EU personal data
Some data comes with legal obligations, not just company preference. For health data in the United States, the key concept is protected health information (PHI): any individually identifiable health information. To handle PHI, Anthropic offers HIPAA-ready services covered by a Business Associate Agreement (BAA), but nothing is covered automatically. On Claude Enterprise, the Primary Owner must activate HIPAA compliance under Data and privacy and accept the BAA. For the first-party API, the Primary Owner must sign the BAA and then ask Anthropic to turn it on.
Even inside a HIPAA-ready organisation, coverage depends on the feature. On Enterprise, chat, projects, artifacts, web search, research and skills are covered. Connectors/MCP, Enterprise Search and Claude in Chrome can be used, but data they send to third parties is not covered by the BAA, and Cowork is not covered at all. Claude Code is not covered under the API's HIPAA readiness, and consumer plans are outside these arrangements entirely. More generally, data processed by third-party tools and integrations falls outside Anthropic's retention arrangements, so each service's own data handling has to be reviewed.
For personal data about EU residents, the sources describe the contractual basis. Anthropic's Data Processing Addendum, with Standard Contractual Clauses, is built into the Commercial Terms of Service, and Anthropic describes its approach to GDPR through its Privacy Policy, DPA and Help Center articles. They do not describe EU data residency: the residency controls documented here offer only "us" and "global" inference. The defensible answer is therefore to use a commercial plan under those terms and route the decision through your organisation's privacy owners. A personal consumer account is not the right place for this data.
| Feature | BAA status |
|---|---|
| Chat, Projects, Artifacts | Covered as Eligible Services |
| File creation & code execution | Covered, excluding network access and use of external websites |
| MCPs / Connectors, Claude in Chrome | Available, but sending data to 3rd parties isn’t covered |
| Cowork | Available, but not covered |
| Claude Design, Claude Slides, Claude Docs [beta] | Not available yet for HIPAA-ready organizations |
Checkpoint 3 of 4· Check yourself
A clinic has just bought Claude Enterprise. A nurse wants to start pasting patient notes into chats the same day. What has to happen first?
BAA coverage is not part of a standard Enterprise plan. The Primary Owner has to activate HIPAA compliance and accept the BAA before PHI is covered.
“Standard Claude Enterprise plans do not include BAA coverage without action from a Primary Owner.”Source: privacy.claude.com
Checkpoint 4 of 4· Exam question
A Claude.ai Pro user wants their conversations excluded from any model training data, but they've enabled the optional "model improvement" setting and use Incognito chats occasionally. What happens to their Incognito chats?
Correct answer: A — Incognito chats are excluded from model training even when the model improvement setting is enabled
- A. Correct. Incognito chats are specifically excluded from model training, even for users who have otherwise opted into the model improvement setting for their other conversations.
- B. Incorrect. The model improvement setting applies to a user's regular conversations; Incognito chats carry their own exclusion regardless of that setting.
- C. Incorrect. The 5-year rule applies to conversations retained under the model improvement setting, which does not apply to Incognito chats since they're excluded from training.
- D. Incorrect. Trust and safety flagging depends on detected policy violations in content, not on whether a chat was conducted in Incognito mode.
Exam traps
Each one states something that sounds right. Open it to see what is actually true.
1.On commercial plans, nothing is ever kept for more than 30 days.Why is that wrong?
Content flagged as violating the Usage Policy is kept much longer: inputs and outputs for up to 2 years and classification scores for up to 7.
2.Any Enterprise admin can set chats to auto-delete after a week.Why is that wrong?
Only a Primary Owner or Owner can set custom retention, and the shortest allowed period is 30 days.
Covered in Enterprise custom retention: who sets it and what it deletes
3.Once an organisation is HIPAA-ready, PHI is covered wherever it goes, including through connectors.Why is that wrong?
Connectors and MCP can still be used, but data they send to third parties falls outside Anthropic's BAA.
Covered in Regulated data: health information and EU personal data
Sources
Every claim above is drawn from one of these pages, quoted as it was written on the date shown.
- 1.https://privacy.claude.com/en/articles/7996866-how-long-do-you-store-my-organization-s-dataOfficial docs
“we automatically delete inputs and outputs on our backend within 30 days of receipt or generation”
↩︎ How long data is kept, and what deletion does“We retain inputs and outputs for up to 2 years and trust and safety classification scores for up to 7 years”
↩︎ Exam trap 1“We retain inputs and outputs for up to 2 years and trust and safety classification scores for up to 7 years”
↩︎ Checkpoint - 2.
“Deleted from our back-end storage systems within 30 days”
↩︎ How long data is kept, and what deletion does - 3.
“Conversation content (your prompts and Claude's outputs) is not retained by default; the exception is Covered Models, which require 30-day retention.”
↩︎ How long data is kept, and what deletion does“Protected health information (PHI) includes any individually identifiable health information.”
↩︎ Regulated data: health information and EU personal data“Data processed by third-party websites, tools, or other integrations is not covered”
↩︎ Regulated data: health information and EU personal data - 4.https://privacy.claude.com/en/articles/10440198-configure-custom-data-retention-controls-for-enterprise-plansOfficial docs
“To set custom retention periods for your organization, you must have either a Primary Owner or Owner role.”
↩︎ Enterprise custom retention: who sets it and what it deletes“Data past its retention period will be permanently deleted and cannot be recovered.”
↩︎ Enterprise custom retention: who sets it and what it deletes“All retention-related actions and changes are automatically tracked in audit logs.”
↩︎ Enterprise custom retention: who sets it and what it deletes“The minimum retention period is 30 days, and each month is counted as 30 days.”
↩︎ Exam trap 2“By default, data is retained indefinitely unless a custom retention period is set.”
↩︎ Prediction - 5.
“The Compliance API lets your organization programmatically pull activity feed events, chat data, and file content across all your Claude deployments.”
↩︎ Enterprise custom retention: who sets it and what it deletes - 6.https://privacy.claude.com/en/articles/8114513-business-associate-agreements-baa-for-commercial-customersOfficial docs
“Standard Claude Enterprise plans do not include BAA coverage without action from a Primary Owner.”
↩︎ Regulated data: health information and EU personal data“Available to use but sending data to 3rd parties via this feature isn’t covered under Anthropic’s BAA.”
↩︎ Exam trap 3 - 7.https://privacy.claude.com/en/articles/7996862-how-do-i-view-and-sign-your-data-processing-addendum-dpaOfficial docs
“Anthropic’s DPA with Standard Contractual Clauses (SCCs) is automatically incorporated into our Commercial Terms of Service”
↩︎ Regulated data: health information and EU personal data - 8.
“Inference geo: Only "us" and "global" are available.”
↩︎ Regulated data: health information and EU personal data