OpenAI's new ChatGPT agent can perform interactive tasks on your behalf

Imagine an AI bot that can fill out online forms, book airline flights, order groceries, and more. That’s the intent of OpenAI’s new Operator, an AI that acts as an independent agent to carry out your commands all on its own.

Also: Operator isn’t worth its $200-per-month ChatGPT Pro subscription yet – here’s why

Released as a research preview on Thursday, Operator is able to interact directly with a web browser. That means it can navigate web pages by typing, scrolling, and clicking in all the right spots, just as you would yourself. The difference here is that Operator aims to do all that without any intervention on your part.

Sounds cool, but Operator is starting off slowly. 

How to try ChatGPT Pro

Beyond its initial research preview status, the tool is now accessible only with ChatGPT Pro subscriptions in the US, which cost $200 a month. As the AI evolves and learns from its mistakes, OpenAI plans to expand its reach to Plus, Team, and Enterprise users and eventually integrate its skills directly into ChatGPT.

ChatGPT Pro users who want to take Operator for a spin should browse to its dedicated web page. Make sure you’re signed in with your OpenAI account. From there, type a request at the prompt just as you normally would with ChatGPT. Only you’ll want to fashion that request as one that asks Operator to carry out tasks on the web independently.

Also: ChatGPT vs. ChatGPT Plus: Is a paid subscription still worth it?

For example, you could ask Operator to find and book a tour of Rome through Tripadvisor, order more bananas and apples from Instacart, or purchase Apple’s AirPods Pro 2 from Amazon. You can even tell the bot to handle several tasks simultaneously. To help Operator carry out actions with specific vendors and websites, OpenAI is working with companies such as DoorDash, Instacart, OpenTable, Priceline, StubHub, Thumbtack, and Uber.

What to expect with Operator

As cool as all this may sound, there are certainly potential pitfalls and problems. Depending on the complexity of the task, Operator could get stuck or make a mistake along the way. In that case, it will try to self-correct. If that doesn’t work, the tool will hand control back to you to intervene.

Operator is also unable to handle confidential information, such as passwords, payment details, and CAPTCHA challenges. If it runs into a website requiring a login or payment card, it will ask you to take over.

Further, Operator will refuse to carry out certain sensitive tasks, such as depositing money online or submitting a job application. It will also ask for approval before completing other types of tasks, such as submitting an online order or sending an email.

Also: Have a genealogy mystery? How I used AI to solve a family puzzle

Privacy is always a concern, especially with AI. To better protect your privacy, you can opt out of training so that Operator won’t use your data for learning. You’re also able to delete all browsing data, log out of all sites, and erase past conversations by going to the Privacy section in Operator’s settings.

Security is another worrisome area. How will cybercriminals and hackers be prevented from exploiting and abusing Operator?

First, the AI is designed to refuse harmful requests and block prohibited content. Second, it will detect and ignore prompt injections in which hackers try to feed malicious information disguised as legitimate requests. Third, a built-in monitor looks out for suspicious behavior and will pause the task if such behavior is detected. Fourth, OpenAI will use both automated and human reviewers to watch for possible threats.

Of course, other privacy and security issues will likely pop up as Operator wends its way through the ChatGPT Pro community. That’s the purpose of limiting its initial use to just Pro subscribers before looking to expand to a wider audience.

Operator uses a new model

To carry out tasks on the web, Operator uses a new model called Computer-Using Agent (CUA), which combines GPT-4o’s vision skills with more advanced reasoning. That combination lets it interact with the menus, controls, and text fields on a web page without the need for a custom API.

Also: The best AI chatbots

“Operator can be asked to handle a wide variety of repetitive browser tasks such as filling out forms, ordering groceries, and even creating memes,” OpenAI said in its news release. “The ability to use the same interfaces and tools that humans interact with on a daily basis broadens the utility of AI, helping people save time on everyday tasks while opening up new engagement opportunities for businesses.”