Story perspectives

OpenAI's Atlas Browser Vulnerabilities Risk User Security

10/30/2025

Summary

Story summary

OpenAI's Atlas browser has vulnerabilities that could let hackers manipulate the AI assistant.
LayerX reports attackers can inject code into ChatGPT's memory, enabling remote execution across devices.
Atlas blocks only 5.8% of phishing attempts, making users up to 90% more vulnerable than traditional browsers because of a CSRF flaw and weak anti-phishing protections.
OpenAI acknowledges the issues and is patching them; experts advise avoiding AI browsers until security improves.

1 of 1

Story summary

OpenAI's Atlas browser has vulnerabilities that could let hackers manipulate the AI assistant.
LayerX reports attackers can inject code into ChatGPT's memory, enabling remote execution across devices.
Atlas blocks only 5.8% of phishing attempts, making users up to 90% more vulnerable than traditional browsers because of a CSRF flaw and weak anti-phishing protections.
OpenAI acknowledges the issues and is patching them; experts advise avoiding AI browsers until security improves.