OpenAI Atlas Browser Vulnerability Lets Attackers Execute Malicious Scripts in ChatGPT

OpenAI’s ChatGPT Atlas browser is described as having a critical vulnerability that lets attackers inject harmful instructions into ChatGPT’s memory and execute remote code, according to LayerX’s findings. The flaw uses a Cross-Site Request Forgery attack to piggyback on a user’s ChatGPT authentication, tainting the memory feature and enabling malicious code deployment during subsequent ChatGPT queries.

LayerX noted Atlas users are by default logged into ChatGPT, meaning authentication credentials remain constantly available for CSRF exploitation, and observed a high real‑world attack success rate across Atlas compared with traditional browsers. In testing, Atlas allowed ninety-seven of one hundred and three attacks to succeed, a failure rate exceeding ninety-four percent, versus about fifty-three percent for Edge and forty-seven percent for Chrome.

The report also warns that tainted memories can cause ChatGPT to generate seemingly legitimate code containing backdoors or remote code execution capabilities, with infection persisting across devices and browsers where the same account is used. The vulnerability was disclosed to OpenAI under responsible procedures, with the article published on 28 October 2025. according to LayerX