ChatGPT Atlas browser launch splits opinion over security and privacy risks
OpenAI's new ChatGPT Atlas browser has launched to considerable fanfare. Beyond ordinary browsing, it includes an agent that can take over the browser to complete tasks on the user's behalf, building on the agent mode the company released in July.
The launch has reignited a debate about how much of a person's digital life should sit under one roof. A single AI browser can touch private email, direct messages, confidential client work, online banking, investments, social accounts and search history. Skeptics acknowledge that the big platforms have tracked users for years, but argue it is still wise to wait for Atlas to mature, especially on security and privacy, before migrating everything from an existing browser.
Security researchers have been warning that AI agents are vulnerable to prompt injection attacks, where invisible instructions embedded in web pages trick an agent into stealing data or taking unwanted actions on the user's behalf. One widely cited assessment holds that the security and privacy risks still feel insurmountably high, and that none of these products should be trusted until independent security researchers have put them through thorough testing.
Enthusiasts counter with three features they call game-changing. An Ask ChatGPT button works on any site, from GitHub to Amazon to Gmail, carrying full page context so there is no more copy-pasting between tabs. Agent mode lets the AI complete tasks autonomously, visibly moving the cursor or working in the background. And browser memory learns preferences, work patterns and habits, letting users search their history by simply describing what they are looking for.
In other Meta news, the standalone Messenger desktop apps for Windows and Mac will shut down on December 15. Users will get an in-app notification when the deprecation process begins and 60 days to keep using the Mac app before being blocked, after which Meta encourages deleting it. Messenger remains available through the Facebook website and messenger.com in any browser.
Meta is also ending the Messenger inbox ad placement: from November 11, campaigns carrying that placement will stop delivering there. Existing campaigns, including those using Advantage+ placements, keep running without interruption and require no action from advertisers.
Meanwhile, Meta's AI translation for Instagram and Facebook Reels has gone live across four languages: English, Spanish, Portuguese and Hindi, with lip-syncing and voice matching that even preserves accents. Where available, a View translations option appears on Reels, and creators can enable the feature in their upload settings.
The launch wave has revived an old critique of Meta: outside of jumping on trends others pioneered, from Stories to short-form video, the company rarely originates a concept that rivals have not already proven. Some industry voices predict smart glasses will overtake the phone as the primary device within a decade, a shift Meta is betting heavily on.
