Is ChatGPT Atlas safe? It's retired, and your browsing data may not be gone

The Morse teamPublished October 26, 20267 min read

No. As of October 22, 2026, ChatGPT Atlas isn't safe to keep using, because OpenAI has retired it: the browser was scheduled to stop working on August 9, 2026, and OpenAI's help center now says Atlas "no longer receives updates." If you used it, check what it left behind. Atlas browser memories sit on OpenAI's side, and OpenAI says deleting your ChatGPT chats doesn't delete them, and uninstalling Atlas may not either.

We read OpenAI's Atlas help center, its release notes, and its prompt injection post on October 22, 2026. OpenAI's help center blocks automated reads, so we read the latest archived copy of each page, captured between June 25 and October 16, 2026, and we name the capture date where it matters. Atlas deserves some credit. OpenAI published unusually concrete limits for it, like a 7-day deletion window for page summaries and a plain admission that prompt injection may never be fully solved.

ChatGPT Atlas key facts, October 2026

Is ChatGPT Atlas still safe to use?

No, because it's a browser nobody is patching. OpenAI's shutdown notice, in its July 2026 version, answered the question itself: "Browsers require ongoing security maintenance, and we do not want users to remain on a discontinued browser that may degrade or stop receiving security updates." The same notice said OpenAI was "moving browser-based agentic capabilities into ChatGPT and Codex."

The archived release notes list no build after March 10, 2026. As far as OpenAI's pages show, Atlas never shipped beyond the Mac. The release notes list only macOS builds, and the enterprise page lists only "Macs with Apple silicon."

Proton's guide, published October 22, 2025, still describes Atlas as a live Mac browser and doesn't mention the retirement, the 7-day summary window, or the diagnostic logs that were on by default. Proton did flag, citing Kaspersky, that AI built into a browser gives malicious pages a direct channel to manipulate it, and OpenAI's own December post bears that out.

What are Atlas browser memories, and how long are they kept?

They're notes ChatGPT kept about the sites you read. OpenAI's data controls page says "web content is summarized on our servers," then filtered to keep out things like government IDs, bank account numbers, and medical records. OpenAI says it deletes "web contents right after they are summarized" and the "privacy-filtered summaries within 7 days." The memories built from those summaries stay until you delete them.

OpenAI's pages disagree on whether memories started on or off. The launch notes say "if you turn on browser memories," and the data controls page says users "opt-in during setup." OpenAI's web browsing settings page says "Browser memories are turned on by default for new users." If you set Atlas up quickly, assume they were on.

Two controls limited what was sent:

• Page visibility. Turning it off for a site meant "ChatGPT can't view the page content, and no memories are created from it."
• On-device summaries. On macOS 26, OpenAI offered to summarize pages on your Mac "so that web contents are not sent to our servers."

Did ChatGPT Atlas train on the pages you browsed?

Not by default, but your Atlas chats followed your ChatGPT setting. OpenAI's data controls page says "Include web browsing" was off by default. It also says that if training was on for your ChatGPT account, it was on for Atlas chats, including "website contents you've attached when using the Ask ChatGPT sidebar, and Browser memories" behind those chats. Our page on whether ChatGPT trains on your data has the opt-out steps.

A separate switch, "Help improve browsing & search," was on by default. OpenAI says it shares diagnostic logs that "can include technical details and publicly known URLs" and doesn't train the model.

Was Atlas agent mode safe with your logged-in accounts?

Only partly, and OpenAI said so. In agent mode, ChatGPT clicked and typed for you, and it could "use sites you're already signed in to," according to OpenAI's agent help page. The launch notes said the agent couldn't run code, download files, read ChatGPT memories, or use saved passwords, and ended: "These efforts don't eliminate every risk."

The risk OpenAI named is prompt injection, where text on a page gives the agent orders. In a December 22, 2025 post, OpenAI showed its own test: a planted email told the agent to send a resignation letter to the user's CEO, and the agent did it while drafting an out-of-office reply. OpenAI wrote that prompt injection "is unlikely to ever be fully 'solved'" and recommended logged-out mode "whenever access to websites you're logged in to isn't necessary."

That advice still applies to the agent features now in the ChatGPT desktop app. Give an agent narrow tasks, and keep it logged out of your bank and email.

Was Atlas incognito private?

From your browser history, yes. From OpenAI, not entirely. OpenAI's data controls page says incognito signed you out of ChatGPT, but "we retain signed-out chats separate from your account for 30 days to detect and prevent abuse," and "Incognito doesn't make you invisible to ChatGPT or the rest of the internet." For more on signed-out use, see our guide to anonymous ChatGPT.

What happens to your Atlas data now?

Each piece lives in a different place, and none of it moved for you. This table draws on OpenAI's retirement page (October 16, 2026 capture) and its data controls page.

OpenAI's retirement FAQ warns: "Do not assume that uninstalling Atlas or changing ChatGPT Memory settings deletes that data." It points requests about Atlas data to the Privacy Portal or OpenAI Support. If you never exported anything, OpenAI says "recovery is not guaranteed."

This isn't legal advice. If a workplace used Atlas, OpenAI's enterprise page said Atlas data "may not be covered by Enterprise data retention" rules and was outside its SOC 2 and ISO attestations, so ask your admin what was kept.

How Morse keeps what you read and ask off company servers

Atlas sent what you read, and what you asked about it, to OpenAI's servers and filed it under your account. A setting decided what happened to it, and the retirement left it there. Morse is a web app, not a browser, and it's built the other way round.

Your chat history is stored in your browser, encrypted, and Morse's servers don't keep your messages or log what you write. Morse doesn't train AI on your content, sell your data, or show ads, so there's no "Include web browsing" switch to check. Clearing your browser, losing your device, or deleting your account deletes your chats, and Morse can't recover them, so keep anything you need.

Then choose a mode by what you're pasting in:

• E2EE, for a page or document no one else should read. Your browser checks the secure enclave's proof that it's genuine, then encrypts the message so only that enclave can open it, and Morse's servers relay text they can't read. It runs open-weight models with no web search, and it rests on the enclave hardware and attestation working as designed. E2EE chats come with Pro, $19 a month. Here's how end-to-end encrypted AI chat works.
• Anonymous, to ask GPT, Claude, or Gemini about a page without the provider knowing it's you. Our proxy removes your account, email, and IP address and sends the message under Morse's own account. The provider still sees the message and may keep it for a limited time to check for abuse, under its own policy, so leave names out of what you paste.
• Private, the default, for everyday questions no host should keep. It runs open-weight models like DeepSeek, Kimi, and GLM, plus Grok. Morse has zero data retention agreements with its inference, search, and other providers, and your identity is stripped as in Anonymous.

Our privacy policy has the details, and our guide to using AI privately covers the rest.

Try a chat from the Morse home page without an account, or start free and paste in the next page you'd have asked Atlas about.