ChatGPT Work Adds Secure Website Sign-Ins

This turns ChatGPT Work from a browser assistant into a genuinely useful action-taking agent, especially for repetitive account workflows. The privacy design is promising, but persistent sessions make permission hygiene and prompt-injection defenses critical.

• –Users enter credentials directly through browser takeover, keeping passwords out of the chat and agent screenshots
• –Persistent cookies let the agent resume future tasks without repeated sign-ins
• –Utilities, appointments, and reimbursements are practical demonstrations of high-value browser automation
• –Agents can still access sensitive account data after authentication, so users should limit permissions and review consequential actions
• –Website compatibility and anti-bot controls remain meaningful constraints on real-world reliability