ChatGPT Work agents can now log into websites while claiming passwords stay hidden
OpenAI updates ChatGPT Work to handle website logins and trigger tasks from Gmail, Slack, and GitHub.
OpenAI announced a new feature in ChatGPT Work that lets AI agents enter websites and handle tasks requiring manual browser navigation, including authentication inputs.
According to the company, ChatGPT Work recognizes login screens and lets users manually input usernames, passwords, and security codes. OpenAI states that credentials are never exposed to the model, used for training, or stored in the chatbot, and users can rely on existing password managers to fill in data. Once authenticated, the session stays active for future tasks, though browser data can be deleted per site or globally in settings.
Supported scenarios include utility setups, appointment bookings, checking insurance expenses, and sorting emails into accounting software. Even after authentication, ChatGPT still requires user confirmation before final actions like bookings or payments. The website login support is rolling out in web and mobile apps for ChatGPT Plus, Pro, and Business users.
Additionally, OpenAI updated scheduled tasks so ChatGPT Work can react to events like new Gmail messages, Slack posts, and GitHub merge requests rather than running only at fixed times. These scheduled tasks are now available for ChatGPT Free users, though free accounts are capped at three active tasks, lack webhook triggers, and can run repeating tasks at most once per day. Previously, OpenAI released an Apple Messages plugin for sorting and editing messages directly from the chatbot.
Source: habr
Comments
This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.