Privacy & the Dorm-Door Rule
One simple rule keeps you safe.
People overcomplicate AI privacy. Here's the rule that handles almost all of it:
Don't paste anything you wouldn't tape to your dorm door.
If you'd hesitate to write it on a sticky note and stick it on your door for the whole hall to see, don't paste it into a chatbot. Easy.
That's the whole privacy framework. Everything else is just examples of what fits in each bucket: and the four toggles inside Claude that govern what happens to the stuff you do paste.
Why this lesson now
In the next few weeks you're going to start putting real things into Claude. Your Voice Profile. Your syllabi. Eventually, your Application Profile with your actual stories and history. Settings tuned now beats settings tuned later, after the stuff is already in.
Privacy at a glance.
Three lights: green, yellow, red.
🟢 Fine to share
- Your first name (or a nickname)
- Year in school, general major
- Class names & topics
- Schedule patterns
- Writing voice & preferences
- General location (city)
- Goals & opinions
- Your own draft writing
🟡 Pause first
- Full legal name
- Home or dorm address
- Specific employer details
- Photos with faces visible
- Specific financial amounts
- Your transcript
- Other people's private info
🔴 Never
- Social Security number
- Credit card / bank numbers
- Passwords or secret tokens
- Medical record numbers
- Student ID number
- Login credentials (school, banking, anything)
- Other students' work
The "other students' work" line is important
If a classmate shares their draft with you for peer review, don't paste it into AI without their permission. That's their writing, their voice, their potential consequence: not yours to share. Same for group-project drafts where someone else wrote a section.
The settings page: and the one toggle that matters most.
The dorm-door rule covers what you paste. The settings page governs what happens to it after.
Getting to Privacy settings
- 1. Open claude.ai in a browser (the desktop / web version is easier; mobile works too).
- 2. Click your profile / initials at the bottom-left of the sidebar.
- 3. Click Settings.
- 4. In the left rail of Settings, click Privacy (sometimes "Data & privacy" or "Privacy controls").
You'll see a few labeled toggles: typically a section for "Help improve Claude" (the training opt-in/out), data retention info, an Export your data button, and Delete account. The key lever on this page is the training toggle.
🟢 Toggle OFF:what happens
- Your chats are not used to train future Claude models.
- Conversations are retained for about 30 days on Anthropic's servers (for safety review and abuse detection), then deleted.
- You can still see and use your chat history normally on your account:"retention" is about Anthropic's servers, not your interface.
- Recommended for: most students, especially anyone who'll paste real essays, real application stories, real personal context.
🟡 Toggle ON:what happens
- Your chats may be used in de-identified form to train future Claude models.
- Conversations are retained for up to 5 years for that purpose.
- Anthropic doesn't sell your data; it stays with Anthropic, encrypted at rest.
- Worth considering if: you actively want to help models get better and you only paste low-sensitivity content.
Chat retention & the emergency drill.
"Retention" is just "how long does this conversation exist." Three things to know: plus a quick drill for when you paste something you wish you hadn't.
Where your chats live, and how to delete
- In your sidebar: as long as you don't delete them, individual chats stay in your sidebar history forever.
- On Anthropic's servers: tied to the training toggle. Off = ~30 days. On = up to 5 years.
- Delete one chat: three-dot menu next to the chat → Delete. Permanent on your end; Anthropic's backend timer (typically 30 days) applies after.
- Delete all chats: Settings → Privacy → Delete all chats or Clear conversations. Useful at end of semester for a clean slate.
- Export first if you want a copy: Settings → Privacy → Export data. Useful if you want transcripts as a process trail (Module 4.5) before a mass delete.
Incognito / temporary chats
Claude's mobile and web apps have a temporary chat mode (sometimes called "incognito"):usually a small icon near the new-chat button. Chats started in this mode aren't saved to your sidebar and aren't used for training regardless of your training toggle. Use it for one-off sensitive questions you don't want lingering in your history. Don't make it your default: you'll lose useful chats: but know it's there.
The sensitive-paste drill
- 1 · Delete the chat. Three-dot menu → Delete. Don't try to "scroll past it" or "edit just the message." Kill the whole chat. New conversation, fresh start.
- 2 · If a credential or password leaked into the chat: change it. Even though Claude isn't actively serving your data anywhere, the password is "out of your direct control" the moment it's pasted into any third-party tool. Same logic as if you accidentally pasted it into a Slack channel.
- 3 · If something serious leaked: like another person's private info, a transcript, an SSN:note the date in your phone's notes app and let the affected person know if it's their info, not yours.
School-issued accounts.
One last thing: and a reminder of why Rule 3 of the Honest Work Code lives here too.
Closing: Claude for Education accounts
- If your school provides Claude through a Claude for Education / Workspace / Enterprise account, the data rules are different. School/enterprise accounts are governed by your school's contract with Anthropic: chats from those accounts are not used for training, and retention follows your school's IT policy.
- The catch: your school's IT department may have administrative access to that account's chats, the same way they can access your school email. Don't paste anything into your school-issued Claude account that you wouldn't put into your school email.
- The clean split: use the school account for school-related work tied to that platform, and your personal Claude account for personal stuff (apps to other schools, internships, your Voice Profile, your Stories Bank). Two accounts, two purposes, no confusion.
Their rules beat your rules: even on privacy
Honest Work Code · Rule 3. A reminder that goes deeper than just academic-honesty policies: schools, employers, scholarship programs, and internship hosts may have specific rules about what AI tools you can use, what data you can put in them, and what counts as a privacy violation in their eyes. A perfectly tuned personal Claude account doesn't override your school's IT policy or your internship's NDA. Read the rules of any new context before you paste; what's fine for your personal study session might be a real problem inside your campus health portal, your employer's system, or your scholarship application portal.
Module 1: what you carry forward
- You have a Claude account on a personal email, with 2FA on.
- You know the five things in the interface (chat, sidebar, attachments, Projects, settings) and you've uploaded a real file.
- Voice mode is set up on your phone, and you know the talk-it-out-then-ask-for-structure move.
- The Dorm-Door Rule is your default. You've made a deliberate choice on what your training toggle is set to. You know the drill if you paste the wrong thing.
You finished Module 1. Up next: 2.1: The Formula.
Account, interface, voice, privacy: all set. Module 2 is the craft of asking:the four-line formula (Act as / Do this / Constraints / Format) that turns "help me with my paper" into a brief Claude can actually act on, with three worked examples.
Continue to 2.1:The Formula →