Are your AI chats private?
What happens to what you type into an AI assistant. None of the big mainstream assistants in this table encrypt your chats so the company can't read them.
Read: your AI chats aren't private →| Company | Trains on your chats by default | Can opt out | Humans may read chats | Hidden from the company |
|---|---|---|---|---|
| Lumo Proton Tight-lipped | No | Yes | No | Partly |
| Leo Brave Discreet | No | Yes | No | Partly |
| Duck.ai DuckDuckGo Discreet | No | Yes | No | Partly |
| Vibe (formerly Le Chat) Mistral AI Chatty | Yes | Yes | Partly | No |
| Claude Anthropic Loose-lipped | Yes | Yes | Yes | No |
| Perplexity Perplexity Loose-lipped | Yes | Yes | Not ruled out | No |
| Copilot Microsoft Loose-lipped | Yes | Yes | Yes | No |
| ChatGPT OpenAI Loose-lipped | Yes | Yes | Yes | No |
| Gemini Google Loose-lipped | Yes | Partly | Yes | No |
| Grok X Oversharer | Yes | Yes | Yes | No |
| Meta AI Meta Oversharer | Yes | No | Yes | No |
| DeepSeek DeepSeek Oversharer | Yes | Yes | Not ruled out | No |
Lumo
How long chats are kept: Proton says it keeps no server-side logs of conversations (data is erased once a reply is generated); saved chat history is synced with zero-access encryption. Guest chats are erased at the end of the session.
Prompts must be decrypted on Proton's servers to generate a reply, so it isn't end-to-end encrypted in the way Proton Mail messages are. The only human review is of anonymized feedback you choose to submit on answers from the Apertus model, which is shared with its university research teams.
Sources (2)
- Lumo privacyProton · Primary sourcearchived
- Lumo security model: How Proton makes AI privateProton · Primary sourcearchived
Leo
How long chats are kept: Not stored by Brave after a response is generated; chat history is kept only on your device.
No account required.
Sources (1)
- Brave Leo AIBrave · Primary sourcearchived
Duck.ai
How long chats are kept: Not stored by DuckDuckGo by default; AI providers must delete prompts immediately after replying, except that OpenAI and Anthropic may cache them in memory for up to an hour, and Anthropic may keep them where required by law or to fight abuse.
Your IP address and identifying metadata are removed before your prompt reaches the AI provider. Only chats you choose to share as feedback are stored and reviewed.
Sources (1)
- How does Duck.ai protect my privacy?DuckDuckGo · Primary sourcearchived
Vibe (formerly Le Chat)
How long chats are kept: Kept until you delete the conversation or your account.
Enterprise is not used for training by default; Team admins can switch training off. Thumbs-up/down feedback is used for training on every plan, and authorised staff can view conversations you report.
Sources (1)
- Can I opt out of my input or output data being used for training?Mistral AI Help Center · Primary sourcearchived
Claude
How long chats are kept: Kept until you delete them. Up to five years (de-identified) if you allow training; otherwise deleted within 30 days of you deleting a chat. Chats flagged for policy violations are kept up to two years.
Existing users were asked to choose via a pop-up whose training toggle was pre-set to on; new users choose during sign-up. Conversations flagged for safety review may be kept for up to two years and may be used to train safety systems even if you opt out.
Sources (3)
- Updates to Consumer Terms and Privacy PolicyAnthropic · 28 Aug 2025 · Primary sourcearchived
- How long do you store my data?Anthropic Privacy Center · Primary source
- How does Anthropic protect the personal data of Claude users?Anthropic Privacy Center · Primary source
Perplexity
How long chats are kept: Kept while your account is active, until you delete it (account deletion completes within 30 days); Incognito sessions expire within 24 hours.
Enterprise accounts are not used for training.
Sources (1)
- Data Collection at PerplexityPerplexity Help Center · Primary sourcearchived
Copilot
How long chats are kept: Stored for 18 months by default unless you delete it. Opting out of training excludes future conversations only.
Work and school (Entra ID) accounts, Microsoft 365 consumer app users, signed-out users and under-18s are not used for training, and nor are users in Brazil, China, Israel, Nigeria, South Korea and Vietnam.
Sources (1)
- Privacy FAQ for Microsoft CopilotMicrosoft Support · Primary sourcearchived
ChatGPT
How long chats are kept: Kept until you delete them; deleted chats are removed within 30 days unless they were already de-identified or OpenAI must keep them for security or legal reasons. Temporary Chats may be kept for up to 30 days.
Business, Enterprise and API data is not used for training by default.
Sources (4)
- Data Controls FAQOpenAI Help Center · Primary sourcearchived
- Temporary Chat FAQOpenAI Help Center · Primary sourcearchived
- Enterprise privacy at OpenAIOpenAI · Primary sourcearchived
- How OpenAI handles data in consumer servicesOpenAI Help Center · Primary source
Gemini
How long chats are kept: Kept for 18 months by default (adjustable). Chats reviewed by humans are kept for up to three years, disconnected from your account, even if you delete them. With activity off, chats are still kept for up to 72 hours.
Turning off 'Keep Activity' (formerly Gemini Apps Activity) stops future chats being used to improve Google's AI unless you send feedback, but also turns off chat history. Temporary Chats are not used for training.
Sources (1)
- Gemini Apps Privacy HubGoogle · Primary sourcearchived
Grok
How long chats are kept: Kept with your account until you delete it; deleted chats and Private Chat conversations are removed within 30 days (unless kept for legal, compliance or safety reasons).
X posts are also used for training unless you turn it off in X's settings.
Sources (1)
- Privacy PolicySpaceXAI · Primary sourcearchived
Meta AI
How long chats are kept: Kept with your account. Chats are also used for ad and content personalisation.
Meta says it won't use sensitive topics (health, religion, politics, sexual orientation, ethnicity) for ads. The ad change did not apply in the EU, UK or South Korea at launch.
Sources (2)
- Improving Your Recommendations on Our Apps With AI at MetaMeta · 1 Oct 2025 · Primary sourcearchived
- Meta greenlights Facebook, Instagram ads based on your AI chatsCNBC · 1 Oct 2025 · Reportingarchived
DeepSeek
How long chats are kept: Stored on servers in China for as long as you have an account (and longer where DeepSeek says it has a legal or business need).
You can opt out of training in DeepSeek's settings.
Sources (2)
- DeepSeek Privacy PolicyDeepSeek · Primary sourcearchived
- South Korea says DeepSeek transferred user data to China and the U.S. without consentCNBC · 24 Apr 2025 · Reportingarchived