OpenAI launches Operator, an AI agent that performs tasks autonomously

OpenAI CEO Sam Altman kicked off this year by declaring in a blog post that 2025 would be a big year for AI agents, tools that can automate tasks and act on your behalf.

Today we are seeing the first real attempt at OpenAI.

OpenAI on Thursday announced the launch of a research preview of Operator, a general-purpose AI agent capable of taking control of a web browser and performing certain actions independently.

The operator is targeting US users first with ChatGPT’s $200 Pro subscription plan. OpenAI says it plans to eventually roll out this feature to more users in its Plus, Team, and Enterprise tiers.

“(The operator) will be (in) other countries soon,” OpenAI CEO Sam Altman said during a livestream Thursday. “Europe will unfortunately take time. »

This initial look at the research is available through Operator.chatgpt.com, but soon OpenAI says it wants to integrate Operator into all of its ChatGPT clients.

The operator promises to automate tasks such as booking accommodation, restaurant reservations and online shopping, according to OpenAI. There are several task categories that users can choose from in the Operator interface, including Shopping, Delivery, Catering, and Travel, all of which allow for different types of automation.

When ChatGPT users enable Operator, a small window appears showing a dedicated web browser that the agent uses to perform tasks, along with explanations of the specific actions the agent is performing. Users can still take control of their screen while Operator works, because Operator uses its own dedicated browser.

OpenAI says Operator is powered by a computer-using agent, or CUA, model that combines the vision capabilities of the company’s GPT-4o model with the reasoning capabilities of OpenAI’s more advanced models. CUA is trained to interact with the front end of websites, meaning it does not need to use developer-facing APIs to access different services.

In other words, the CUA can use buttons, navigate menus, and fill out forms on a web page, much like a human would.

OpenAI claims to work with companies like DoorDash, eBay, Instacart, Priceline, StubHub, and Uber to ensure the operator complies with these companies’ terms of service.

“The CUA model is trained to ask the user for confirmation before finalizing tasks that have external side effects, e.g. before submitting an order, sending an email, etc., so that the user can double-check the model’s work before it becomes permanent.” OpenAI writes in documents provided to TechCrunch. “(This) has already proven useful in a variety of cases, and we aim to extend this reliability to a wider range of tasks.”

But OpenAI cautions that CUA is not perfect. The company says it “does not expect CUA to perform reliably in all scenarios at this time.”

Out of an abundance of caution, OpenAI also requires oversight for certain tasks, such as banking transactions, that the CUA and operator might perform primarily on their own. Users will have to take over to enter their credit card information, for example. OpenAI says the operator does not collect or capture any data.

“On particularly sensitive websites, such as email, Operator requires active user supervision, ensuring that users can directly detect and correct any potential errors the model might make,” OpenAI says in its support documents.

This limits Operator’s usefulness, of course, but also ensures that the agent isn’t hallucinating and, say, spending your mortgage payment on accent chairs. Google has taken a similar approach with its Project Mariner AI agent, which also doesn’t fill in information like credit card numbers.

A future agent

OpenAI has been rather slow to develop an AI agent compared to its competitors (see: agents from Rabbit, Google and Anthropic), which may have something to do with the security risks associated with the technology.

When an AI system can take actions on the web, it opens the door to much more dangerous use cases from nefarious actors. You can automate AI agents to orchestrate phishing scams or DDoS attacks, or have them grab concert tickets before anyone else can. Especially for a tool as widely used as ChatGPT, it’s important that OpenAI takes steps to prevent these kinds of exploits.

OpenAI seems to think that Operator is safe enough to release in its current form, at least as a research preview.

Operator is OpenAI’s boldest attempt yet to create an AI agent. Last week, OpenAI released Tasks, giving ChatGPT simple automation features such as the ability to set reminders and schedule prompts to run at a set time each day.

The tasks gave ChatGPT users familiar, but necessary, features to make ChatGPT as convenient to use as Siri or Alexa. However, Operator presents capabilities that the previous generation of virtual assistants could never offer.

AI agents have been touted as the next big thing in AI after ChatGPT: a new technology that will change the way people use the Internet and their PCs. Instead of just providing and processing information, agents can, in theory, take actions and actually do things.

With the release of the first concrete version of OpenAI agents, it will soon become clear how realistic this vision is.