ChatGPT Work is becoming considerably more useful for tasks that previously required users to take over and log in themselves.
OpenAI is expanding Work’s ability to interact with websites through a cloud browser, including workflows that require access to authenticated services. The change could make ChatGPT far more capable of completing everyday tasks end-to-end rather than simply preparing instructions for the user.
Basically: if it’s something you’d normally have to open a browser and click through yourself, try asking ChatGPT Work to do it.
Logging into websites is rolling out today on web and mobile for Plus, Pro, and Business users. pic.twitter.com/PlPhrCIGes
— ChatGPT (@ChatGPT) August 25, 2026
ChatGPT Work can now handle website logins
The biggest change is the ability to get past one of the most important barriers for AI browser agents: the login screen.
Instead of stopping when a website asks for a username and password, users can provide their credentials through a secure login flow. The goal is to let Work continue the task without requiring users to repeatedly take control of the browser.
OpenAI’s cloud-browser documentation describes the remote browser as a way for ChatGPT to carry out delegated web tasks, while its newer browser capabilities distinguish between remote cloud browsing and the built-in browser, which can also support sign-in and password management.
The important part is that the password itself isn’t intended to become part of the normal ChatGPT conversation. Users enter sensitive credentials through the appropriate secure interface rather than simply pasting them into a prompt.
That distinction matters because giving an AI agent access to an account is fundamentally different from asking it to browse a public website.
From answering questions to actually getting things done
This is where ChatGPT Work starts to look less like a chatbot and more like a personal digital assistant.
For example, a user could potentially ask Work to:
- File or manage an insurance claim
- Reorder household or business supplies
- Check an account for a specific update
- Complete repetitive forms
- Use information from an uploaded image to identify items that need restocking
- Continue a multi-step workflow across websites and connected apps
The broader idea is simple: tell ChatGPT what needs to happen, rather than explaining every individual click.
That fits with OpenAI’s original positioning of Work as an agent designed for longer, multi-step jobs. Work can use connected apps and files, create finished documents and other deliverables, and continue working while the user monitors its progress and approves important actions.
Security remains a critical part of the experience
Giving an AI access to logged-in websites obviously raises a much bigger security question than ordinary chatbot use.
OpenAI has been building controls around agentic browsing, including permission prompts and confirmation requirements for consequential actions. Its documentation says ChatGPT is designed to request confirmation before actions that could create financial, legal, account or other real-world commitments.
That means the goal isn’t necessarily to let an AI blindly submit everything it encounters.
Instead, users can delegate the repetitive parts while retaining control over actions that matter.
For anyone using Work with sensitive accounts, the same basic rule still applies: check the website, account and requested action before approving it.
Work is also becoming more proactive
The website-login capability isn’t the only change making ChatGPT Work more interesting.
OpenAI has also expanded event-triggered scheduled tasks, allowing Work to respond to activity in connected services such as Gmail, Slack and GitHub.
For example, a task can be configured to react when a new Gmail message arrives, when a monitored Slack channel receives a message, or when supported GitHub pull-request activity occurs.
This changes the model from:
“Ask ChatGPT to do something.”
to:
“Let ChatGPT watch for something and do the work when it happens.”
That’s a much more powerful concept for an AI assistant.
Free users get more scheduled-task access
OpenAI is also expanding scheduled-task availability beyond paid users.
Current documentation says Free users can create scheduled tasks, although with tighter limits than paid plans. Free users can schedule a one-time task or a recurring task no more than once per day, while paid plans receive more flexible scheduling and higher task limits.
There is an important distinction, however: event-triggered Work tasks are not currently available to Free and Go users. Those require eligible paid plans such as Plus, Pro, Business, Enterprise or Edu.
Why this matters
The bigger story isn’t simply that ChatGPT can now log into another website.
It’s that several capabilities are starting to come together:
Memory + connected apps + browser access + scheduled triggers + website actions.
Together, they create something much closer to a persistent digital assistant.
Instead of opening ChatGPT every time a task needs to be done, users can increasingly give Work an objective and let it handle the workflow, returning only when it needs information, permission or confirmation.
That could eventually make traditional “AI assistant” features such as reminders and simple automation feel surprisingly limited.
The next phase of AI assistants
AI agents have spent the last few years getting better at understanding what people say. The next challenge is getting them to reliably do something about it.
ChatGPT Work is moving in that direction.
The ability to interact with authenticated websites is particularly significant because much of the internet sits behind login screens. If AI agents can securely access those services while keeping users in control of sensitive actions, an enormous number of previously manual workflows become candidates for automation.
For now, website compatibility, security controls and approval requirements will still determine what Work can actually complete. But the direction is clear: ChatGPT is increasingly being designed not just to answer questions, but to take responsibility for completing the work.
For continuing updates on OpenAI, ChatGPT, Codex, GPT models, and AI industry developments, bookmark:
