OpenAI ChatGPT Cloud Browser Now Supports Secure Website Logins
OpenAI has updated the cloud browser in ChatGPT Work to allow users to log into websites by entering credentials and 2FA codes directly through a secure form. The credentials are sent straight to the remote browser, meaning the underlying ChatGPT model cannot access them, and sessions are maintained via isolated cookies. This update is a significant step for agentic workflows, enabling AI agents to bypass login walls and perform tasks on private or subscription-based websites. It balances automation capabilities with security by keeping sensitive credentials hidden from the AI model itself. To prevent abuse, an additional review model checks login requests for phishing risks, and critical actions like payments still require manual user confirmation. Users can manage site access permissions and clear their saved session cookies at any time through the cloud browser settings.
## BACKGROUND
ChatGPT Work utilizes a cloud-operated browser running on OpenAI's remote servers to research and interact with web pages. This setup is similar to Remote Browser Isolation (RBI), a cybersecurity model that executes browsing activity in a virtual environment to protect the user's local device from web-based threats.