OpenAI ChatGPT Work Can Now Sign In and Act for You

OpenAI is giving ChatGPT Work a more consequential kind of autonomy. According to The Verge's report, the workplace agent can now sign in to websites and complete tasks such as booking a DMV appointment, cancelling a reservation, or filling out a job application, without OpenAI seeing the user's login credentials.

The update is a clear shift from assistants that generate instructions to agents that execute work inside the browser. It also makes the permission boundary more important: users are delegating actions, not merely asking for text.

What changed

ChatGPT Work is designed to handle tasks that require navigating websites and interacting with forms. The key promise is credential isolation. OpenAI can provide the agent's reasoning and orchestration layer without receiving the passwords used to access the sites where the work happens.

The feature arrives as AI companies compete to make agents useful beyond coding and research. OpenAI has described Work as a way to connect AI to the software employees already use, while the company continues expanding its workplace agent strategy. That direction is also visible in the broader move toward persistent team context explored in n8n Lab's analysis of OpenAI workspace agents.

Why builders should care

For automation teams, browser access expands the surface area of possible workflows, but it also raises the cost of mistakes. A useful implementation needs explicit approvals for sensitive actions, scoped sessions, audit logs, and a clear fallback when a website changes its interface.

The practical lesson is straightforward: browser automation should be treated as a tool with production controls, not as an unrestricted extension of the model. Credential isolation reduces one class of risk, but it does not stop an agent from submitting the wrong form or making an irreversible booking.

OpenAI's move signals where workplace agents are heading next: from answering questions about a company's systems to operating those systems directly, with security and human oversight becoming part of the product rather than an afterthought.