ChatGPT Work's remote-browser authentication

ChatGPT Work can now authenticate with login-gated websites on a user's behalf while keeping credentials entirely with the user. When the agent reaches a login prompt, it pauses and hands control to the user, who enters credentials directly into a secure remote-browser form; the agent then resumes the task. Session cookies persist across sessions, so users need to sign in only once per site. OpenAI states the agent never types the user's password, stores credentials, or invokes a password manager, and inputs entered during the manual takeover step are not collected or seen by the model. Screenshots are disabled during the takeover step. The feature covers web and mobile and extends the agent's reach beyond the public web to authenticated dashboards, SaaS tools, internal portals, paywalled content, and multi-step dealings with landlords, insurance companies, and government agencies. Website policies, authentication checks, and human review remain limits on what the agent can accomplish.