General · 24 October 2025
OpenAI's ChatGPT Atlas Browser Faces Major Security Risks Despite Built-in Safeguards
OpenAI's new ChatGPT Atlas browser has been found vulnerable to prompt injection attacks within hours of its launch, with cybersecurity experts warning of significant security risks. The Chromium-based browser integrates AI agent capabilities, allowing autonomous web browsing and task completion, but researchers have demonstrated successful attacks including clipboard injections and data exfiltration attempts. Despite extensive red-teaming and safety measures, OpenAI's security chief acknowledges prompt injection remains an unsolved frontier problem.
Reported by The Decoder · How we write briefs