AI privacy myths: eight things people believe about their chats, checked

The Morse teamPublished October 25, 20265 min read

The most common AI privacy myths treat a setting with a private-sounding name as if it kept your words away from the AI company. Incognito windows, VPNs, paid plans, training opt-outs, and the delete button each limit one thing, and none of them stops ChatGPT, Claude, or Gemini from receiving and storing what you wrote.

We checked eight beliefs against the companies' own help pages and policies and one court filing, all read on October 20, 2026. Each myth below gets a short correction with its source and a link to the page where we cover it in depth. Google's Chrome help deserves credit here: it says in plain words that Incognito "doesn't make you invisible."

Does incognito mode hide your AI chats?

No. A private browser window only stops your browser from keeping a record on your device, and the AI company still receives every message you send.

Google's Chrome help says that after an Incognito session ends, "Chrome doesn't retain site data or a record of the sites you visited," and that Incognito "doesn't make you invisible": the sites you visit and "organizations that manage your network" may still see your activity (Google, read October 20, 2026). That describes your laptop, not OpenAI's servers. If you sign in to ChatGPT inside the window, the chat is a regular one, and OpenAI says regular chats "remain saved to your account until you delete them" (OpenAI). Closing the window doesn't touch that copy.

The word "incognito" causes a second mix-up. Chrome's Incognito is a browser feature. The incognito or temporary chats inside ChatGPT, Claude, and Gemini are settings on the company's side, and each company still keeps those chats for a set period. Anthropic, for one, says Claude's incognito chats "are retained for either 30 days (default)," or longer under an Enterprise organization's custom retention setting (Anthropic). Using one doesn't switch on the other. We list what each app's temporary chat keeps in how to use AI privately.

Does a VPN make ChatGPT private?

No. A VPN changes the IP address OpenAI sees, and OpenAI's privacy policy says it also collects your name, contact details, payment information, and device identifiers (OpenAI, read October 20, 2026). The same policy says "We don’t 'sell' Personal Data," and uses personal data to personalize the ads Free and Go users see on ChatGPT, subject to their settings. We compare the VPN with every other method in anonymous ChatGPT.

Are paid AI plans kept out of training?

No, not personal plans. Anthropic's training page covers Claude Free, Pro, and Max alike and excludes only commercial products such as Claude for Work and the API (Anthropic, updated March 16, 2026). We read 15 apps' rules in which AI companies train on your data.

Does turning off training stop staff from reading your chats?

No. OpenAI lists four reasons authorized staff may access content, including abuse investigations, support, and legal matters, and the training opt-out removes only the fourth, model improvement (OpenAI, read October 20, 2026). The allegations about OpenAI's reviewers are in ChatGPT Project Lily.

Is a deleted ChatGPT chat gone right away?

No. OpenAI schedules a deleted chat "for permanent deletion from its systems within 30 days," with exceptions for de-identified copies and security or legal holds (OpenAI, read October 20, 2026). For the months in 2025 when a court made OpenAI keep deleted chats, see the ChatGPT court order.

Are AI chats privileged like a lawyer's?

No. In United States v. Heppner, Judge Jed Rakoff held that a defendant's chats with Claude weren't privileged, writing that "No such relationship exists, or could exist, between an AI user and a platform such as Claude" (memorandum, filed February 17, 2026). Anthropic's own guidance, dated August 26, 2026, says the ruling "turned on the consumer privacy terms and the absence of attorney direction" (Anthropic). This isn't legal advice, and the other cases are in can AI chats be subpoenaed.

Does an open-source model mean your chats are private?

No. Open weights describe the model, not where it runs, so the privacy of a chat depends on whose computer runs it and what that operator's policy says.

DeepSeek, GLM, and Qwen are all open-weight, and the myth assumes that makes every way of using them private. DeepSeek's own app is covered by a privacy policy that says "we directly collect, process and store your Personal Data in People's Republic of China" and that it uses personal data to "train and improve our technology" (DeepSeek, last updated February 10, 2026). On your own laptop, the same weights keep the chat on that machine unless the software you run them with sends it somewhere, and on a third-party host they follow that host's rules. The gap between DeepSeek's app and its weights is in is DeepSeek safe.

Does end-to-end encryption mean the AI can't see your prompt?

No. The model has to read your prompt to answer it, so in AI chat end-to-end encryption means only a sealed enclave running the model can decrypt the message, as Brave's post on its encrypted Leo chats describes (Brave, published October 15, 2026). We explain who checks the enclave in end-to-end encrypted AI chat.

How Morse keeps your chats from depending on a setting's name

In every myth above, your chat sits on a company's server, under your account, in a form the company can read. A setting can change what the company does with that copy, but the copy still exists. Morse is designed so that no readable copy sits on its servers, and you pick the privacy mode per chat.

• E2EE, for the prompt you'd want no one to read. Your browser checks the enclave's cryptographic proof itself before anything leaves your device, then seals the message so only that enclave can open it. Morse's servers relay ciphertext they can't read, and every E2EE reply has a lock you can open to see each check that passed. Third-party providers run the enclaves, not Morse, and the guarantee rests on the enclave hardware and attestation working as designed.
• Private, the default, so the model's host keeps no copy for a delete button to miss. Open-weight models like DeepSeek, Kimi, and GLM, plus xAI's Grok, run under the zero data retention agreements Morse has with its inference, search, and other providers, so they keep nothing. Where a provider confirms zero retention on each reply, Morse checks it and refuses a reply that doesn't; otherwise Morse relies on the provider's commitment.
• Anonymous, for what a VPN can't do with GPT, Claude, or Gemini. Our proxy removes your account, email, and IP address and sends the message under Morse's own account, so the provider gets the question without knowing it's you. It may keep the message for a limited time to check for abuse, under its own policy, so leave identifying details out of the text.

Some things hold in every mode. Morse doesn't train AI on your chats, sell your data, or show ads, whatever plan you're on. Your history is stored in your browser, encrypted, and locked when you sign out, and a temporary chat is kept in memory only. Clearing your browser, losing the device, or deleting your account deletes your chats, and Morse can't recover them, so keep a copy of anything you need. Because Morse's servers don't keep your messages, we usually have very little to give authorities, as our privacy policy says.

Try a chat from the Morse home page with no account, or start free. End-to-end encrypted chats come with Pro, at $19 a month.