Key facts
- ChatGPT Work's browser can now complete tasks on websites requiring user sign-in.
- Users enter credentials once, and the agent can continue working.
- Sessions may remain signed in for future tasks, reducing the need for repeated logins.
- OpenAI asserts that the model cannot see or store usernames and passwords.
- The agent can act on the account after successful authentication.
- Users can clear browsing history and sessions per site via Settings.
OpenAI has introduced a new agentic browser capability for ChatGPT Work, enabling the AI to log into websites that require user credentials. This feature, detailed in the August 25 release notes, allows users to enter their login information once, after which ChatGPT can proceed with tasks and maintain an active session for future use. OpenAI states that the model does not see or store usernames and passwords, and these are not utilized for training purposes. The design allows the agent to act on the account after authentication, with sessions persisting until the user manually clears their browsing history. While this offers convenience by eliminating repetitive logins for tasks like filing forms or pulling statements, it also presents a potential security trade-off, as the agent gains access to an account that would typically require user presence. OpenAI notes that safeguards are in place for password security, but the ongoing session access remains a point of consideration. The feature is currently live on both web and mobile versions of ChatGPT Work, with options to clear sessions individually per site available in the settings.