OpenAI allowing some users to test a new ChatGPT feature that uses it artificial intelligence to operate a web browser to book flights, buy groceries, find bargains, and do many other tasks on the internet.

The new tool, called Operator, is an AI agent: It relies on an AI model trained on both text and images to interpret commands and learn how to use a web browser to execute it. OpenAI claims that it has the potential to automate many daily tasks and routine work tasks.

OpenAI Operator follows the opposite release of the two Google and Anthropic, which has are displayed able to use the web. AI agents widely seen as the next stage of evolution for AI is following chatbots, and many companies are jumping on the hype train by touting them. In most cases, they are very limited in their abilities and simply use a modeling language to automate things that are normally done with regular software.

“AI is evolving from this tool that can answer your questions to one that can also move the world, executing complex, multistep workflows,” said Peter Welinder, OpenAI’s VP of product. “We’re going to see a huge impact on people’s productivity—but also the quality of work that people can do.”

OpenAI admits that giving ChatGPT access to a web browser introduces new risks, and it says that the Operator can sometimes behave badly. It says that it is implementing various new protections and plans to extend the capabilities of the Operator gradually.

Welinder and Yash Kumar, product and engineering lead for OpenAI’s Computer Using Agent, say the plan is to learn from how people use the tool. They acknowledge that the tool can make unwanted bookings or purchases but add that a lot of work has been done to ensure it asks questions before taking any risks. “It will come back to me and ask for confirmations before taking steps that may be irreversible,” Kumar said.

OpenAI has now also released a new “system card” outlining problems the Operator may encounter. This includes the potential for it to misunderstand commands or deviate from what a user is asking; to be misused by users; or to be targeted by cybercriminals.

“It also poses an incredible safety challenge,” Kumar said. “Because your attack vector area and your danger vector area are getting higher and higher.”

The operator will initially be available as a “research preview” for ChatGPT users with a Pro account, which costs a whopping $200 per month. The company says it plans to expand access while rolling out the tool gradually, because it will inevitably make some mistakes along the way.

In several demonstrations, Operator showed the potential for AI to take a more active role as a web assistant. The tool has a remote web browser and a chat window for communicating with a user.

At WIRED’s request, the Operator was asked to book an Amtrak train trip from New Haven, Connecticut, to Washington, DC. It went to the right website and entered the necessary information correctly to bring up the timetable, then asked for further instructions. If a user is logged into Amtrak’s website or a browser profile with stored credit card information, the Operator can go ahead and book a ticket—even if it’s designed to request a ticket first. permission.

Kumar asked Operator to book a table at Beretta, a restaurant in San Francisco. The program went to the OpenTable website, found the right restaurant, and searched for availability before asking what to do next. OpenAI says it has partnered with many popular sites, including OpenTable, to ensure Operator works seamlessly with them.

The new tool is based on OpenAI’s GPT-4o AI model, which can see a browser and web page and communicate with typed text. The tool includes additional training designed to help it understand how to carry out tasks online. OpenAI will also make the Computer Use Agent available through its API.