Lock down your AI assistants

Every major AI assistant keeps some record of what you type, and out of the box the settings lean toward the vendor. ChatGPT's personal plans have model training switched on until you turn it off,¹ Claude's memory is on by default for consumer plans,² and Google keeps Gemini activity for 18 months unless you change it.³ Half an hour with the settings puts you back in charge of what is kept, what is remembered and what helps train the next model.

1. 1

   Step 1: List every AI assistant you use

   Include the obvious ones (ChatGPT, Claude, Gemini, Copilot, Meta AI) and the built-in ones in your email, phone, browser, office apps and messaging apps. Note whether each is signed in with a personal or a work account. Lock down the personal ones below, and hand the work ones to whoever manages your business accounts.

   How you know it worked

   You have a short list showing each assistant, the account it uses, and whether that account is personal or work.

2. 2

   Step 2: ChatGPT: turn off training

   Open Settings, then Data controls, and switch off Improve the model for everyone; it starts switched on for personal Free, Go, Plus and Pro accounts.¹ Voice mode has its own sharing control, so check that too. Skip the thumbs-up and thumbs-down buttons on sensitive chats, because feedback can send the whole conversation for training even after you opt out.¹

   How you know it worked

   Data controls shows Improve the model for everyone switched off, on the website and in the phone app.

3. 3

   Step 3: ChatGPT: review memory, shared links and temporary chats

   Under Settings, then Personalization, then Memory, review what ChatGPT remembers and delete anything sensitive; deleting a chat does not delete a memory saved from it.⁴ Under Data controls, open Shared links and delete the ones you no longer need, because anyone holding a link from a personal account can read that chat.⁵ For one-off sensitive questions, use a Temporary Chat: it skips history, memory and training, though OpenAI may keep it for up to 30 days for safety.¹

   How you know it worked

   Memory holds only details you are comfortable keeping, and the Shared links list is empty or holds only links you still need.

4. 4

   Step 4: Claude: choose your training setting

   Open Settings, then Privacy, and set the Help improve our AI models toggle the way you want.⁶ If it is on, Anthropic may keep de-identified copies of your chats for up to five years in its training pipeline.⁷ Deleting a chat removes it from Anthropic's systems within 30 days and keeps it out of future training,⁸ and incognito chats (the ghost icon) are never used for training.⁷⁹

   How you know it worked

   Settings, then Privacy, shows the training choice you made.

5. 5

   Step 5: Claude: review memory

   Memory is on by default for Free, Pro and Max accounts, and by default it leaves out sensitive topics such as health and beliefs.² In Settings, then Memory, you can pause memory or reset it, which permanently deletes it. Memories created from a chat stay even after you delete that chat, so remove those memories individually.²

   How you know it worked

   Settings, then Memory, shows memory paused, or it holds nothing you would mind a stranger reading.

6. 6

   Step 6: Gemini: set Keep Activity, auto-delete and memory

   On gemini.google.com, open Settings & help, then Activity. Keep Activity is on by default for adults; turn it off, or leave it on and shorten auto-delete from the default 18 months to 3 months.¹⁰³ Memory sits under Settings & help, then Personal Intelligence, and only works while Keep Activity is on.¹¹

   How you know it worked

   The Activity page shows Keep Activity off, or an auto-delete period you picked.

7. 7

   Step 7: Microsoft Copilot: check memory and personalization

   In the updated Copilot app released in August 2026, open Settings, then Personalization.¹² Turn off Saved memories if you don't want Copilot remembering details, and use Manage, then Delete all memories, because switching the toggle off does not erase what is already saved.¹² The same page has toggles for One shared experience (sharing with Bing, Edge and MSN) and ad personalization.

   How you know it worked

   Personalization shows Saved memories, One shared experience and Allow ads personalization set the way you chose.

8. 8

   Step 8: Meta AI: delete chats and limit group-chat sharing

   In the Meta AI app, open Menu, then Settings, then Data & privacy, then Manage your information, where you can delete all chats and media.¹³ Meta says it uses AI interactions to improve its AI, and announced that from December 16, 2025 it would also use them to personalize content and ads in most regions; it describes no opt-out from using AI chats for ads, and you can only adjust ads in Ads Preferences.¹³¹⁴ In Messenger, anyone who uses Meta AI in a chat can share recent messages with it, so turn message sharing off in sensitive group chats.¹⁵

   How you know it worked

   History in the Meta AI app is empty, and message sharing with Meta AI is off in your family and work group chats.

9. 9

   Step 9: Review connected apps and what they can do without asking

   In ChatGPT, connected apps are under Settings, then Plugins (or Apps); disconnect what you don't use and set permissions to Always ask or Allow read actions rather than Allow all actions.¹⁶¹⁷ On Claude Team and Enterprise plans, owners can set connector tools to Needs approval or Blocked; on personal plans, review your connectors and disconnect the ones you don't use.¹⁸ In Gemini, open Customize, then Connectors.¹⁹

   How you know it worked

   Each assistant lists only apps you actively use, and where your plan allows it, none of them can send, share or delete without asking you first.

10. 10

    Step 10: Protect the account itself

    Your chat history is a detailed record of your life or your work, so guard the login like your email login. Turn on multi-factor authentication (MFA) or a passkey for each AI account and for the Google, Microsoft or Apple account behind it. Sign out of AI apps on old phones and shared computers.

    How you know it worked

    Each AI account asks for a second factor or a passkey when you sign in on a new device.

11. 11

    Step 11: Business: move work into business accounts and set the defaults

    ChatGPT Business and Enterprise, Claude Team and Enterprise, Gemini with Google Workspace, and Microsoft Copilot with a work account all say they keep your data out of model training by default.^20212223 Business plans add admin controls; custom data-retention periods are generally an Enterprise-tier feature.²⁰²⁴ Owners should decide which connectors staff may use and switch off feedback buttons where the plan allows, since feedback can send whole chats to the vendor; Claude Team and Enterprise owners, for example, can turn off Rate chats.²¹

    How you know it worked

    Everyone who uses AI for work signs in with a company account, and the admin console shows connector and feedback settings you chose on purpose.

12. 12

    Step 12: Set a reminder to check again

    Vendors move these settings and change their terms, sometimes with little notice. Put a recurring reminder in your calendar to repeat this walk-through every six months, and whenever an assistant announces a privacy update.

    How you know it worked

    A recurring calendar reminder exists for your next check.

• We checked every setting on the vendors' own help pages on the review date shown on this page. Menu names move often; if a label doesn't match, search the vendor's help center for the setting's name.
• Turning training off is not deletion. Data already used in a training run stays in the models trained on it.⁷
• Temporary and incognito chats are still kept for a short safety window, and chats flagged by safety systems can be kept much longer.⁷
• Turning off Gemini's Keep Activity doesn't mean nothing is kept. Even with it off, Google keeps chats for 72 hours, and chats already read by human reviewers can be kept for up to three years.³
• Copilot's training settings depend on the app version. Microsoft says prompts, responses and files in the updated app aren't used to train the core models behind its AI;²⁵ in the older app, the training switches sit under Privacy.²⁶
• Disconnecting an app stops future access but doesn't delete past chats that already used that data.¹⁶
• Business plans have gaps too. ChatGPT Business keeps share links inside your workspace but has no company-wide switch to turn sharing off,⁵ and Microsoft notes that Copilot's web searches go to Bing under separate terms from your business data.²³
• Settings can't control what other people share, such as a relative who brings AI into a group chat or a colleague who runs your email through an assistant.
• Features and settings vary by country, age and account type. We describe what adult US personal accounts saw when we checked.
• This guide covers five major assistants. AI built into other apps and devices has its own settings, so check each vendor's privacy page.

• ThreatHomeBusiness

  AI chatbot data leaks and shadow AI

  Whatever you paste into an AI assistant can be stored, read by reviewers or used to train future models, depending on the app, the plan and your settings. In a 2025 survey of workers in 47 countries, 48% of those who use AI at work said they had uploaded sensitive company or customer information into public AI tools. The fix is mostly unglamorous: business accounts for work, the right settings at home, and a clear rule about what never goes in.

  Read
• ThreatHomeBusiness

  Prompt injection and AI agent hijacking

  When an AI assistant reads your email, files or web pages, it can mistake text written by a stranger for instructions from you. That is prompt injection, and when the assistant can also send, buy or delete things, an attacker can steer it. The UK's National Cyber Security Centre (NCSC) says it may never be fully fixed, so the safe move is to limit what your AI can reach and keep yourself in the approval loop.

  Read
• GuideBusiness

  A one-page AI use rule for your team

  Your team is probably using AI already, so the useful question is which tools, with what data, and who checks the result. This guide helps you write those answers on one page, get everyone to sign it, and review it twice a year. When you need more, our full AI Acceptable Use Policy generator uses the same structure and is free with a free account.

  Read
• GuideHome

  Digital hygiene at home: the household baseline

  A safe household runs on a few habits: one password manager, passkeys or multi-factor authentication (MFA) on the accounts that matter, automatic updates, and backups you've actually tested. AI adds one more: a family safe word and a callback rule, so a panicked call that sounds like family gets checked before anyone pays. People 60 and older reported $7.7 billion in losses to the FBI in 2025, so bring older relatives into the plan.

  Read
• ExplainerHomeBusiness

  Who owns your AI conversations?

  In practice, the company that runs the assistant holds your conversations, and its terms decide who can read them, whether they train future models and how long they're kept. Some consumer plans, including ChatGPT's, use your chats for training unless you switch it off, while the major business plans say they don't by default. You can usually export or delete your chats, but deleting doesn't undo training that already happened, and legal holds can override deletion.

  Read