Metro asked me for a quick reaction when OpenAI launched Atlas, its ChatGPT-branded web browser, and the comparison I reached for on the spot has stuck with me since: OpenAI is parking its tanks on Google's lawn. I think that's still the right way to read what's actually going on here.
What Atlas actually is
Atlas isn't a conventional browser with a chatbot bolted on, it's closer to the reverse: a browser-shaped wrapper around ChatGPT. There's no traditional address bar, you type what you want into ChatGPT's own interface, and an "Ask ChatGPT" sidebar sits alongside whatever page you're on so you can query it directly, summarise an article, get context on a hotel, check whether a product fits what you're looking for. There's also an early, premium-only agent mode that can carry out tasks on your behalf, from booking a restaurant table to handling a shopping list, plus an opt-in "browser memories" feature that carries context from your past searches into future ones. For now it's macOS only, with Windows and mobile versions planned, and OpenAI has been candid that it's still early.
Why I don't think this is just a browser launch
The reason I gave Metro that particular line is that ChatGPT, used as an answer engine, already competes with the thing that made Google Google: being the default first stop when you want to know something. A browser is just the next logical layer, if people are increasingly asking a chatbot instead of typing into a search box, then owning the browser itself is the obvious way to make that the default behaviour rather than a habit users have to opt into. Agentic AI, systems that don't just answer questions but actually go and do things on your behalf, is where most of the serious investment in this industry is currently pointed, and if it works the way its backers hope, it changes how people work and live, not just how they search.
The trade-off
The privacy questions that came with the coverage are worth taking seriously rather than waving away. A tool that can see your bookmarks, your saved passwords and your full browsing history in order to be useful is, by definition, collecting a lot about you, and the "free" part of a free assistant is rarely free in the way it looks. OpenAI's own position is that most of the relevant settings, including whether your browsing trains its models, default to off, and that sensitive categories like financial or medical information are filtered out of what gets remembered. That's a reasonable starting position, but reasonable defaults from one company today don't substitute for users understanding what they're trading away, and the security specialists quoted alongside the launch made a fair point: the best protection is still to be deliberate about what personal, financial or family information you let any assistant, browser-shaped or otherwise, get anywhere near.
Read the full piece on Metro>>