What you will be able to do
- Recognise the categories of data that call for a pause before you paste or upload them into Claude
- State when a consumer (Free, Pro, Max) account's chats may be used for model training, and what turning the setting on or off actually changes
- Explain why commercial plans (Claude for Work, the API, Claude Gov) differ by default, and name the one user action that can still send a conversation to Anthropic for training
Key concept
Your plan decides the data rules — Whether your conversations can train Claude, and how long they are kept, depends on whether you are on a consumer plan or a commercial one. Anthropic publishes the rules for the two on separate pages, and the answers are different, so find out which plan you are on before you judge whether something is safe to share.
1.Pause before you paste: what counts as sensitive
The privacy decision happens before Claude sees anything, at the moment you paste text or attach a file. Once content is in a conversation, the retention and training rules for your account apply to it. So the first skill is to recognise what you are holding.
Anthropic's own guidance for consumer users names the categories to be careful with. It recommends being thoughtful about sharing financial information (SSNs, card numbers, bank details), health records or medical information, passwords and login credentials, and confidential business or personal documents. Those four categories line up with the kinds of data most workplace policies care about: personal data about individuals, regulated data such as health or financial records, secrets that grant access, and confidential commercial material such as unreleased results, contracts or deal documents.
How the content got into the conversation also matters. For training purposes, Anthropic does not count raw content that Claude reads through a connector such as Google Drive or an MCP server as part of the chat. Content you copy from those sources directly into the conversation is treated differently. In other words, once you paste a document into the chat, it becomes conversation content and is handled like everything else you typed.
| Category Anthropic names | Typical workplace example |
|---|---|
| Financial information (SSN, credit card numbers, bank account details) | A customer payment export or payroll file |
| Health records or medical information | An employee sick-leave note or patient record |
| Passwords or private login credentials | An API key or shared service-account password |
| Confidential business or personal documents | A draft merger agreement or board pack |
Checkpoint 1 of 5· Check yourself
An employee connects Claude to a shared drive and asks it to summarise a folder. They then copy one full contract from that drive into the chat. For consumer model-training purposes, which content counts as part of the conversation?
Raw connector content is excluded from the training data, but anything copied directly into the conversation becomes part of it.
“though data may be included if it’s directly copied into your conversation with Claude.”Source: privacy.claude.com
Checkpoint 2 of 5· Exam question
A Claude.ai Free plan user deletes a conversation from their chat history. According to Anthropic's consumer data retention policy, when is that conversation removed from Anthropic's back-end storage systems?
Correct answer: A — Within 30 days of the deletion request, per Anthropic's standard back-end purge window
- A. Correct. Deleted conversations are removed from chat history immediately, then purged from Anthropic's back-end storage systems within 30 days.
- B. Incorrect. Deletion is immediate only from the visible chat history; back-end systems still take up to 30 days to fully purge the data.
- C. Incorrect. The 5-year window applies to data retained under the optional model improvement setting, not to deleted conversations that were never opted in.
- D. Incorrect. The 2-year window applies specifically to inputs and outputs flagged for trust and safety violations, not to routine user-initiated deletions.
2.Consumer plans: the user chooses, and the choice is not retroactive
On Claude Free, Pro and Max, and when those accounts use Claude Code, the user controls whether their chats help improve Claude. The setting is under Settings > Privacy, labelled "Help Improve our AI models". Exam questions on this topic usually turn on timing: which chats does the setting affect?
If model improvement is on, Anthropic may keep the data in de-identified form for up to five years in its model-training pipelines. That applies only to new chats, or old chats that are resumed, after the setting was switched on. Turning it on does not reach back into earlier history.
Turning it off has a wider effect. Anthropic will stop using both previous and new chats and coding sessions in future training. Deleting a single chat also keeps it out of future models. Neither action can undo what has already happened: data that is already in a training run in progress, or in a model that has already been trained, stays there. That is why the safest moment to keep confidential material out of training is before it is pasted, not afterwards.
There are two exceptions to keep in mind. Incognito chats are never used to improve Claude, even when model improvement is on. In the other direction, conversations flagged by safety classifiers may still be used to improve Anthropic's internal trust-and-safety models, whatever the setting says.
Checkpoint 3 of 5· Check yourself
A Max user has model improvement turned on. Which of their chats will never be used to improve Claude?
Incognito chats are excluded even when model improvement is on. Flagged chats, by contrast, may still be used for trust-and-safety work.
“Your Incognito chats are not used to improve Claude, even if you have enabled Model Improvement in your Privacy Settings.”Source: privacy.claude.com
Checkpoint 4 of 5· Exam question
A Claude.ai Pro user's conversation is flagged by automated safety classifiers, which generate safety classification scores in addition to the conversation content. How long may Anthropic retain those safety classification scores compared to the flagged inputs and outputs?
Correct answer: A — Safety classification scores may be retained for up to 7 years, longer than the 2-year window for flagged inputs and outputs
- A. Correct. Under the consumer retention policy, flagged inputs and outputs may be retained up to 2 years, while safety classification scores may be retained for up to 7 years.
- B. Incorrect. This reverses the actual durations; safety classification scores are the ones retained longer, up to 7 years, not deleted immediately.
- C. Incorrect. The 5-year figure applies to the optional model improvement and feedback retention windows, not to trust and safety flagged content or its scores.
- D. Incorrect. The 30-day window applies to standard back-end deletion of routine deleted conversations, not to safety classification scores from flagged content.
3.Commercial plans: off by default, with a feedback exception
Commercial products are Claude for Work (Team and Enterprise), the Anthropic API and Claude Gov. On these, inputs and outputs are not used for training by default. There is no individual toggle to check, because the default is already the protective one.
The exception is feedback. If someone explicitly reports feedback or a bug, for example with the thumbs up/down button, Anthropic stores the entire related conversation for up to five years. The feedback is de-linked from user and customer IDs, and it may be used for research and training. So on a work account, one click on a rating button can send a confidential conversation outside the commercial default. To prevent this, a Primary Owner or Owner of a Team or Enterprise plan can turn off the Rate chats setting under Organization settings > Data and Privacy.
The commercial relationship also changes who is in charge of the data. Under the Commercial Terms, the customer organisation is the controller of the data its users submit. Anthropic acts as a processor and handles that data only as instructed. In practice, your organisation's rules decide what you may share, and Anthropic's guidance tells commercial users to consult their organisation's administrators about Claude usage policies.
| Question | Consumer (Free, Pro, Max) | Commercial (Claude for Work, API, Claude Gov) |
|---|---|---|
| Used for training by default? | Only if the user allows model improvement | No |
| Retention for training if allowed | De-identified, up to 5 years, new or resumed chats only | Not applicable by default |
| Thumbs up/down feedback | Conversation stored up to 5 years | Conversation stored up to 5 years; Owners can disable Rate chats |
| Incognito chats | Never used to improve Claude | Not used for training by default |
Checkpoint 5 of 5· Match them up
Match each situation to how training applies
Tap a term, then the definition that fits it.
Commercial plans start with training off, and explicit feedback is the exception that Owners can disable. Consumer plans depend on the user's model-improvement setting.
“If you explicitly report feedback or bugs to us (e.g. via our thumbs up/down feedback button)”Source: privacy.claude.com
Exam traps
Each one states something that sounds right. Open it to see what is actually true.
1.Turning on model improvement on a consumer account makes all of that user's past chats available for training.Why is that wrong?
The up-to-five-year training retention covers only new chats, or old chats that are resumed, after the setting is switched on.
Covered in Consumer plans: the user chooses, and the choice is not retroactive
2.Switching model improvement off removes your data from Claude models that have already been trained.Why is that wrong?
Turning it off stops future use only. Data already in training runs in progress, or in models already trained, stays there.
Covered in Consumer plans: the user chooses, and the choice is not retroactive
3.On an Enterprise or API account, nothing a user does can ever send a conversation to Anthropic for training.Why is that wrong?
Explicit feedback, such as the thumbs up/down button, stores the whole conversation for up to five years and may be used to train models.
Covered in Commercial plans: off by default, with a feedback exception
Sources
Every claim above is drawn from one of these pages, quoted as it was written on the date shown.
- 1.
“we also recommend being thoughtful about sharing highly sensitive personal details such as:”
↩︎ Pause before you paste: what counts as sensitive - 2.
“It does not include raw content from connectors (e.g. Google Drive), including remote and local MCP servers”
↩︎ Pause before you paste: what counts as sensitive“though data may be included if it’s directly copied into your conversation with Claude.”
↩︎ Checkpoint - 3.
“we may retain your data in a de-identified format for up to 5 years in our model training pipelines”
↩︎ Consumer plans: the user chooses, and the choice is not retroactive“we will not use your previous or new chats or coding sessions for future model training”
↩︎ Consumer plans: the user chooses, and the choice is not retroactive“This retention period only applies to new or resumed chats after the model training setting has been enabled.”
↩︎ Exam trap 1“Your data will still be included in model training runs that are already in progress, or in models that have been trained.”
↩︎ Exam trap 2“This retention period only applies to new or resumed chats after the model training setting has been enabled.”
↩︎ Prediction“Your Incognito chats are not used to improve Claude, even if you have enabled Model Improvement in your Privacy Settings.”
↩︎ Checkpoint - 4.https://privacy.claude.com/en/articles/12109829-how-do-i-change-my-model-improvement-privacy-settingsOfficial docs
“If our safety classifiers flag your conversations, they may still be used to improve our internal trust and safety models”
↩︎ Consumer plans: the user chooses, and the choice is not retroactive - 5.
“By default, we will not use your inputs or outputs from our commercial products”
↩︎ Commercial plans: off by default, with a feedback exception“you can disable the ability for members of your organization to submit feedback to Anthropic via the thumbs up / down button”
↩︎ Commercial plans: off by default, with a feedback exception“By default, we will not use your inputs or outputs from our commercial products”
↩︎ Key concept“store the entire related conversation, including any content, custom styles, conversation preferences, or model settings, in our secured back-end for up to 5 years”
↩︎ Exam trap 3“If you explicitly report feedback or bugs to us (e.g. via our thumbs up/down feedback button)”
↩︎ Checkpoint - 6.https://privacy.claude.com/en/articles/9267385-does-anthropic-act-as-a-data-processor-or-controllerOfficial docs
“Anthropic only processes the data as instructed by the customer in order to provide the Claude service.”
↩︎ Commercial plans: off by default, with a feedback exception“You should consult with your organization's administrators regarding Claude usage policies.”
↩︎ Commercial plans: off by default, with a feedback exception