OpenAI is shutting down Atlas, the browser agent launched in October 2025, after just eight months. Agent browsing features are shifting to the updated Chrome extension and ChatGPT desktop app. The company says it’s integrating what it learned from Atlas and user feedback. Existing users are receiving transition notifications.

The new Computer Use feature in the desktop app lets ChatGPT handle background tasks: clicking, typing, moving files, across apps and browsers, as a one-off action or recurring task. The Chrome extension runs ChatGPT directly in your browser sidebar.

Why this matters to you. When Atlas launched, it looked like an attempt to challenge Chrome. Now OpenAI is consolidating: everything in ChatGPT, no separate browser. For you using AI in daily work, that means fewer tools to open and agent capabilities living where you already work. The company is choosing integration over replacement.

But OpenAI is giving up on collecting browsing data at scale—an advantage that stays with Google. Atlas joins the list of OpenAI products that closed or lost momentum: plugins, some apps, the standalone Agent, the Sora model. The pattern is clear: the company experiments, shuts things down fast, and moves features into the main product.

The Details

What came before. Atlas debuted in October 2025 as a standalone browser with agent capabilities: it could navigate, fill forms, run web tasks on your behalf. It was a direct move at Chrome, built on the idea of creating a browser designed for AI from the ground up instead of bolting AI onto a traditional one. Eight months later, OpenAI declares sunset and announces the features are going elsewhere.

Where Atlas is going. Agent browsing capabilities move to two places:

• The updated Chrome extension, which brings ChatGPT to your browser sidebar. You browse Chrome, ChatGPT sees the page and can act on it.
• The new Computer Use feature in the desktop app, which lets ChatGPT control your system: open apps, click, type, move files. Works in any browser and any app, as a one-time task or recurring automation.

Sources confirm OpenAI is integrating lessons from Atlas and collected feedback, but they don’t detail which specific features survive and which don’t. Atlas users are getting a change notification, though sources don’t specify an exact sunset date.

Why this signals something. Pulling Atlas after eight months means OpenAI isn’t betting on a proprietary browser anymore. The strategy becomes: stay where users already are (Chrome, the desktop app), don’t ask them to switch tools. For the end user, that’s more convenient: one assistant that works everywhere. For OpenAI, it’s a trade-off: without its own browser, it doesn’t collect the browsing data that gives Google competitive advantage.

The pattern of closed products. Atlas joins a growing list: ChatGPT plugins (shut down), some experimental apps, the Agent as a standalone product (never took off), the Sora video model (underused). The Decoder flags this pattern explicitly. OpenAI launches products, kills them fast when they don’t gain traction, and reabsorbs features into the core.

What this means for builders. If you build on OpenAI tools, this closure pace matters. Features stay, but the access point changes: an integration that relied on Atlas needs rewriting for the Chrome extension or Computer Use. Consolidation into ChatGPT reduces the surface to track, but it also means OpenAI won’t hesitate to retire young products if they don’t work out.

What we don’t know. Sources don’t say if some Atlas features won’t migrate, or if there are technical limits to the Chrome extension (for example, whether it can do everything Atlas did or if some actions are off-limits). There’s no exact Atlas closure date or details on migration paths for production users. When a company kills a product this young, it usually means real usage was below expectations: OpenAI hasn’t shared numbers.

Computer Use is different. The new feature operates at the OS level, not just the browser. It’s the next step in the agent idea: instead of a closed environment (Atlas browser), the agent sees and controls the whole desktop. This opens wider use cases (automations touching multiple apps), but brings bigger security risks: an agent with system access can cause damage if it makes a mistake. Sources don’t dive into how OpenAI manages permissions and limits.