OpenAI Introduces ChatGPT Atlas: A Chromium-based browser with a built-in AI agent

OpenAI just launched ChatGPT Atlas, a new AI browser that embeds ChatGPT at the core of navigation, search, and on-page assistance. Atlas is available today for Free, Plus, Pro, and Go users, with a Business beta and Enterprise/Edu opt-in; Windows, iOS, and Android builds are “coming soon.”

What ChatGPT Atlas is?

Atlas is a Chromium-based browser that keeps a persistent ChatGPT interface in the new tab page and as an “Ask ChatGPT” sidebar on any site. Users can summarize pages, compare products, extract data, and edit text in-place (cursor-level assist in form fields). Atlas also introduces optional ‘Browser memories’ that retain privacy-filtered summaries of pages you visit to personalize later assistance.

A preview “agent mode” lets ChatGPT take actions in your browser: opening tabs, clicking, and completing multi-step tasks (e.g., research + shopping) with explicit user approval checkpoints. The agent runs with hard boundaries: it cannot run code in the browser, download files, install extensions, access your filesystem, or read saved passwords/autofill; pages it visits in agent mode are not added to history.

Key launch facts

• Engine & base: Atlas is ‘built on Chromium.’
• Platform: macOS first (Apple Silicon, macOS 12+), other platforms planned.
• Import: passwords, bookmarks, and history can be imported from other browsers.
• Privacy defaults: content you browse is not used to train models unless you opt in; a separate toggle (‘Help improve browsing & search’) shares diagnostics and is on by default. Incognito signs you out of ChatGPT; signed-out chats are retained separately for 30 days to prevent abuse.

How Atlas compares to Google Chrome

What’s better than Chrome (as of now)?

• Native AI agent and sidebar: ChatGPT is first-class. The sidebar and in-field editing operate on any page; agent mode can execute tasks across tabs with user-visible controls. Chrome requires add-ons or external apps for equivalent agentic behavior.
• Task-centric new tab and unified results: Atlas’ new tab blends chat with search links, images, videos, and news, reducing context switching.
• Browser memories (optional): privacy-filtered, time-bounded summaries that improve future assistance; on-device summarization is available on newer macOS builds. Chrome lacks equivalent feature integrated with a conversational model.
• Agent safety rails clearly documented: explicit prohibitions (no code execution, no file downloads, no extension installs, no password/autofill access) and ‘logged-out’ agent mode reduce blast radius when delegating tasks. Chrome has no built-in web agent requiring such guardrails.

What’s the same as Chrome?

• Rendering stack and core UX: Being Chromium-based, Atlas inherits modern web compatibility, tabbed browsing, password/passkey manager, and familiar settings/menus; bookmarks and data import mirror Chromium conventions.
• Incognito semantics: private windows exclude activity from history and ChatGPT account context (Atlas signs you out in Incognito), analogous to Chrome’s private mode separation.

What’s worse than Chrome (at launch)?

• Platform coverage: Atlas is macOS-only today; Chrome is cross-platform (desktop/mobile). Windows/iOS/Android for Atlas are planned but not shipping yet.
• Enterprise maturity: Business is beta; Enterprise/Edu require admin enablement. Chrome’s enterprise controls are long-standing.
• Extensions/devtools posture: Documentation does not state Chrome Web Store compatibility, and Atlas’ agent explicitly cannot install extensions. OpenAI lists ‘improved developer tools’ on the roadmap, suggesting parity gaps with Chrome’s mature DevTools ecosystem. Treat extension support as unconfirmed at launch.
• Telemetry default: ‘Help improve browsing & search’ diagnostics are on by default (separate from training opt-in). Chrome also collects diagnostics by default, but the Atlas setting is a new surface that teams must audit.

ChatGPT Atlas meaningfully upgrades the browser into an AI-native workspace: persistent ChatGPT surfaces (new tab, sidebar, in-field edits) reduce context switches for summarization, comparison, and extraction; a preview Agent mode coordinates multi-step tasks across tabs; and optional Browser memories and clear data controls (training opt-in off; diagnostics toggle on) are documented.

• Good: Chromium-level compatibility and easy migration (import passwords, bookmarks, history) plus explicit safety boundaries for the Agent.
• Bad: macOS-only at launch, extension/devtools parity with Chrome remains unstated, and the Agent cannot install extensions or download files—limiting automation scope compared to Chrome’s advanced ecosystem.

Michal Sutter is a data science professional with a Master of Science in Data Science from the University of Padova. With a solid foundation in statistical analysis, machine learning, and data engineering, Michal excels at transforming complex datasets into actionable insights.