Is ChatGPT Atlas safe to use?
OpenAI’s retired Chromium-based AI browser. ChatGPT sat beside the page you were reading; optional browser memories could keep facts from sites you visited; and agent mode could click around, including on sites where you were signed in. Atlas has been retired and no longer receives updates. This card covers the former browser, not the current ChatGPT chat product.
Last reviewed: 12 October 2026
Does ChatGPT Atlas train on what you type?
Yes, unless you switch it off — and you can.
Can you turn ChatGPT Atlas's training off?
Atlas is retired, so its in-browser controls are no longer a current opt-out path. For Atlas-specific data, do not assume that deleting ChatGPT chats, uninstalling Atlas, or changing ChatGPT Memory deletes browser memories, cookies, or other Atlas data. If you still need saved Atlas data, contact OpenAI Support before deleting local browser data; recovery is not guaranteed. If you previously exported passwords, import them into a supported password manager and then delete the readable CSV. Do not share cookie or session exports. For the ChatGPT conversations that remain separate from Atlas browser data, open Settings → Data controls and turn off Improve the model for everyone. You can instead select Do not train on my content in OpenAI’s Privacy Portal. Either choice prevents new eligible personal ChatGPT conversations from being used to improve models. Feedback is a separate exception: if you submit thumbs-up or thumbs-down feedback, OpenAI says the associated conversation may be used for training. Use Temporary Chat for conversations you do not want added to history or used for training; OpenAI says Temporary Chats are deleted after 30 days and may be reviewed only for abuse.
Is what you put into ChatGPT Atlas private?
Atlas stored ordinary browser data such as history, cookies, and site data. Its Incognito mode kept the session out of local browser history and did not link it to your ChatGPT account, but it did not hide activity from your internet provider, school, employer, or the sites you visited. OpenAI said signed-out chats were retained separately for 30 days to detect and prevent abuse. Browser memories added another layer. When they were on, Atlas summarized page content on OpenAI’s servers and kept selected facts or insights as memories you could view, archive, or delete. OpenAI said it deleted the page content after summarizing it and deleted privacy-filtered summaries within seven days. It described filters intended to exclude IDs, bank details, credentials, addresses, medical records, and financial information, and said some sensitive sites were blocked from summarization. Those filters were a safeguard, not a guarantee: Axios reported that EFF testing described by The Washington Post found Atlas had memorized reproductive-health queries and a real doctor’s name; OpenAI said that was not intended. Page visibility could stop ChatGPT from reading a site, adding it to browser memories, or using it in chat responses. Training used two controls. Include web browsing was off by default, so the pages you browsed were not used to train models unless you turned it on; pages that opted out of GPTBot stayed out even then. Atlas chats, page content attached through the sidebar, and browser memories used in those chats followed the ChatGPT account’s Improve the model for everyone setting. OpenAI says Business and Enterprise content is not used for training by default. A separate Help improve browsing & search control, on by default, sent diagnostic details and publicly known URLs; OpenAI described that as product diagnostics rather than model training. Agent mode could act on sites where you were already logged in. OpenAI said it paused on some sensitive sites, could not run code or access other apps or the file system, and offered a logged-out mode that did not reuse existing cookies. OpenAI also warned that hidden instructions on a webpage or in email could still manipulate the agent and that its safeguards would not stop every attack. Atlas has been retired and no longer receives updates. Bookmarks, open tabs, and browser history did not transfer automatically. Users must sign in to websites again in a supported browser and should not share Atlas cookie or session exports. ChatGPT conversations remain separate from Atlas browser data and follow the ChatGPT account’s settings. Deleting ChatGPT chats, uninstalling Atlas, or changing ChatGPT Memory settings does not delete Atlas browser memories, cookies, or other Atlas-specific data.
So, is ChatGPT Atlas safe to use?
Mostly, with care. We currently rate ChatGPT Atlas "Use With Caution" (privacy risk: Medium). Worth reading the points above before you share anything personal.
Our take
“A helper in the browser sounded useful. Then it started summarizing pages on someone else’s servers. Memories stayed off. The bank tab stayed in another browser. Atlas is gone; I still check ChatGPT’s Data controls.”
The official documents behind this review
We read these so you don’t have to. Check them yourself any time.
Wondering how we decide these ratings? Read how we review AI privacy.
Knowing what’s safe is half of it.
The other half is knowing what to do with these tools. We teach that part: tutorials, prompts, plain English, no judgment. $7 a month, and the first seven days are free.
Show Me How This Works →Start free. Cancel anytime. No judgment, ever.