Citenook1Blog · Privacy
Is Claude safe? What Anthropic does with your chats
Yes, Claude is generally safe to use. Anthropic encrypts your chats, says its employees can't read them by default, and doesn't sell your data. The catch is training: on the Free, Pro and Max plans, Anthropic uses your chats to train its models if "Help improve our AI models" is on, and keeps them up to 5 years when it is. Team, Enterprise and the API don't train on your data by default.
This guide covers whether Claude trains on your data and how to turn it off, how long chats are kept, who can read them, memory, using Claude for work, how it compares with ChatGPT and Gemini, and other risks. Facts come from Anthropic's own privacy and help pages, checked on 16 October 2026. Citenook isn't connected to Anthropic.
What "safe" means here
People asking "is Claude safe" usually mean one of five things:
- Is my data private? Mostly yes, with the training setting to check. That's most of this guide.
- Is it safe for work files? Business plans are; personal plans need care. See using Claude for work.
- Is the app real? Yes, if you get it from Anthropic. See other risks.
- Is the AI itself safe? Anthropic publishes a safety policy and reports on misuse it stops.
- Is it safe for teens? Claude's own terms require users to be 18 or older; more below.
Does Claude train on your data?
It depends on your plan and one setting. Until 2025, Anthropic didn't train on consumer chats. On 28 August 2025 it changed its consumer terms for Free, Pro and Max, and asked every existing user to choose by 8 October 2025 whether their chats could be used for training.
- The setting: Settings, Privacy, "Help improve our AI models." It covers your chats and Claude Code sessions on a personal account.
- If it's on: new and resumed chats can be used for training, kept in a de-identified form "for up to 5 years."
- If you turn it off: Anthropic says it "will not use your previous or new chats" for future training. That's broader than ChatGPT, where switching off only covers new chats. Models already trained, or training runs already in progress, keep what they used.
- Rating a reply with thumbs up or down sends that conversation as feedback, which Anthropic keeps for 5 years, even with training off.
Anthropic's pages don't say which way the switch starts for new accounts, so check yours.
How long Claude keeps your chats
| What | How long (Free, Pro, Max) |
|---|---|
| A chat you delete | Gone from your history right away, from Anthropic's systems within 30 days |
| Chats with training on | Up to 5 years, de-identified, in training pipelines |
| Thumbs up or down feedback | 5 years |
| Chats flagged for breaking the usage policy | Up to 2 years; safety scores up to 7 years |
| Incognito chats | 30 days; never used for training |
In every case, Anthropic may keep chats longer when the law requires it or to resolve disputes.
Who can read your Claude chats?
Anthropic says: "By default, Anthropic employees cannot access your conversations." There are two exceptions: when you share a chat as feedback, and when a chat needs review under the usage policy, in which case only designated Trust and Safety staff see it, on a need-to-know basis. Data is encrypted in transit and at rest. That isn't end-to-end encryption, so Anthropic's systems can still process it.
Anthropic's privacy policy says it doesn't "sell" your personal data, as the law defines it.
Memory: what Claude remembers about you
- On by default for Free, Pro and Max on the web, desktop and mobile apps. Team and Enterprise owners decide for their organization.
- Deleting a chat doesn't delete its memories. Anthropic says related memory entries "won't be removed" when a conversation is deleted; delete them yourself in Settings, Memory.
- Some things are never saved, even if you ask: government ID numbers, criminal history, financial account numbers and immigration status.
- Incognito chats don't feed memory and don't use it.
Is Claude safe for work and confidential files?
On a business plan, mostly yes. Anthropic says that "by default" it won't use inputs or outputs from its commercial products, meaning Team, Enterprise and the API, to train its models. API data is deleted within 30 days. Organizations can apply for zero data retention, though Anthropic still keeps its safety classifier results. Its privacy center lists ISO 27001, ISO 42001 and SOC 2 Type I and II.
| Plan | Price (16 October 2026) | Trains on your chats |
|---|---|---|
| Free | $0 | If you allow it |
| Pro | $20 a month, or $17 a month billed yearly | If you allow it |
| Max | From $100 a month | If you allow it |
| Team | $25 per seat a month, $20 billed yearly | No, by default |
| Enterprise | $20 per seat a month plus usage, 20 seats minimum | No, by default |
On a personal plan, turn training off and use incognito for anything sensitive. Anthropic itself advises care with financial details, health records and confidential business or personal documents. If you handle client files, your professional rules still apply on any plan; our guides to ChatGPT for lawyers, ChatGPT for accountants and AI for HR cover those rules, and they apply to Claude too.
Claude vs ChatGPT vs Gemini on privacy
| Claude | ChatGPT | Gemini | |
|---|---|---|---|
| Personal plans train on chats | If "Help improve our AI models" is on | Yes, unless you turn it off | Yes, with Keep Activity on (default for adults) |
| Turning it off covers old chats | Yes, for future training | No, new chats only | Not stated |
| How long chats are kept | Until you delete them, then within 30 days | Until you delete them, then within 30 days | Auto-deleted after 18 months by default; reviewed chats up to 3 years |
| Staff can read chats | Not by default | Authorized staff and contractors | Human reviewers, including providers |
| Business plans train | No, by default | No, by default | No (Workspace) |
For ChatGPT in detail, see is ChatGPT confidential. The plan matters more than the brand.
Other risks to know about
- Fake apps and sites. Security researchers have reported fake Claude download sites and ads. Get the apps from claude.com/download or the official app stores, and don't click download ads.
- Claude in Chrome. Anthropic warns that web pages can try to trick the browser agent (prompt injection), and that it can see whatever is on screen, so keep it away from banking and other sensitive sites.
- The AI itself. Anthropic publishes a Responsible Scaling Policy (version 3.4, July 2026) and regular threat reports on misuse it has stopped; the latest covers December 2025 to August 2026.
- Teens. Anthropic's consumer terms require users to be at least 18. Common Sense Media, an outside group, rates Claude "Moderate risk" for teens and says they shouldn't use it for mental health support.
How to use Claude more safely
- Check the training switch: Settings, Privacy, "Help improve our AI models."
- Use incognito for anything sensitive. It skips memory and training, though chats are kept 30 days.
- Review memory in Settings, Memory, and delete what it shouldn't keep.
- Don't rate replies in chats you want kept out of feedback.
- Manage shared chats and uploaded files in Settings, Privacy, and delete old ones.
- Use Team, Enterprise or the API for work data, and remove names before pasting.
When files shouldn't leave your computer
For files you can't send to any company, the AI has to run on your own machine. Citenook is a Mac app that answers questions from your documents with a model running on the Mac, so nothing is uploaded, and every answer links to the passage it came from. If you'd rather have Claude write the answers, Citenook can use your own Anthropic API key: then the API terms apply (no training by default), only the passages picked for that answer are sent, and names and numbers are hidden first.
It's Mac only, doesn't browse the web or write code like Claude, and starts at $9.99 a month after a 30-day free trial. See privacy and security for what goes where.
Questions
Is Claude safe to use?
For most people, yes. Chats are encrypted, employees can't read them by default, and Anthropic doesn't sell personal data. Check whether "Help improve our AI models" is on, and use incognito for sensitive chats.
Does Claude train on my data?
On Free, Pro and Max, only if "Help improve our AI models" is on in Settings, Privacy. On Team, Enterprise and the API, not by default. Feedback you send with thumbs up or down can be used either way.
How do I stop Claude from training on my chats?
Open Settings, Privacy, and switch off "Help improve our AI models." Anthropic says it then won't use your previous or new chats for future training.
Can Anthropic employees read my chats?
Not by default. Exceptions are chats you send as feedback and chats flagged for usage policy review, which only designated Trust and Safety staff can see.
Is Claude incognito private?
More private: incognito chats aren't used for training or memory and don't stay in your history. Anthropic still keeps them for 30 days.
Is Claude safer than ChatGPT?
On personal plans, Claude's defaults are a little more private: staff can't read chats by default and switching training off also covers old chats. On business plans, neither trains on your data by default.
Sources, checked 16 October 2026: Anthropic consumer terms update, consumer terms, how long Anthropic stores data, model training, changing the training setting, how Anthropic protects data, commercial products, zero data retention, certifications, memory, incognito chats, Claude in Chrome safety, privacy policy, Responsible Scaling Policy, pricing and downloads; Common Sense Media AI risk assessment. ChatGPT and Gemini facts from our ChatGPT confidentiality guide. Claude screenshot from our own account settings.