OpenAI ships the Atlas browser

OpenAI released ChatGPT Atlas, a web browser built on Chromium with ChatGPT embedded throughout the interface rather than bolted on as an extension. A sidebar assistant could answer questions about the page being viewed, summarise content and rewrite selected text; an “agent mode,” available in preview to Plus, Pro and Business subscribers, could carry out multi-step tasks directly on websites, such as booking a reservation, comparing flights or assembling a project brief. OpenAI described agent mode as an early, imperfect capability likely to make mistakes on complex workflows. The browser launched first for macOS only, with Windows, iOS and Android versions promised as coming soon.

Atlas was widely read as a direct move against Google’s dominance of the browser market, since owning the browser gives a company first claim on user intent and search-style queries — the same logic that had made Chrome central to Google’s advertising business. It followed a wave of AI-native and AI-augmented browsers from other companies, including Perplexity’s Comet, intensifying competition over who would own the interface between users and the open web as chat interfaces increasingly substituted for traditional search and browsing.

The launch drew scrutiny almost immediately: security researchers demonstrated within days that Atlas’s agent mode could be hijacked through prompt injection, a vulnerability class that would recur across the emerging category of browser agents through the rest of 2025.