Particle.news
Download on the App Store

OpenAI’s ChatGPT Atlas Browser Faces Security Tests as Company Accelerates Updates

Agentic browsing requires elevated account access, raising new privacy risks that researchers say current defenses do not fully solve.

Overview

  • Atlas is live on macOS with Windows, iOS, and Android versions promised, positioning OpenAI’s AI-first browser as a new gateway to the web.
  • Security researchers demonstrated indirect prompt-injection exploits against AI browsers, with Brave calling the issue systemic and OpenAI’s CISO conceding it remains an unsolved problem.
  • OpenAI published a short-term roadmap through product lead Adam Fry that includes user profiles, tab groups, an opt-in ad blocker, a model picker, and faster, more reliable agent behavior.
  • Android Headlines reports planned agent refinements and deeper integrations, including quicker responses, a steadier pause state, and connections to services like Google Drive and web-based Excel.
  • Early reviews find the Ask ChatGPT sidebar helpful yet clunky and the agent slow, and experts advise limiting agent access, using logged-out mode when possible, and treating copied content as a potential risk.