OpenAI launches Operator, an AI agent that performs tasks autonomously

OpenAI CEO Sam Altman started this year in a blog post saying that in 2025 Be great for AI agentstools that can automate tasks and take action on your behalf.

Now we see the first real attempt at OpenAI.

OpenAI announced Thursday that it is launching a research review of Operator, a general-purpose artificial intelligence agent that can take control of a web browser and perform certain actions independently.

The carrier is coming to US users first on ChatGPT’s $200 Pro subscription plan. OpenAI says it plans to eventually roll this feature out to more users at the Plus, Team, and Enterprise tiers.

“(Operator) will be available in other countries soon,” OpenAI CEO Sam Altman said during a live stream Thursday. “Europe, unfortunately, will take some time.”

A preview of this early research is available via operator.chatgpt.com, but soon OpenAI Operator says it wants to integrate it into all ChatGPT clients.

According to OpenAI, the operator promises to automate tasks such as booking travel places, restaurant reservations and online shopping. There are several task categories that users can choose from in the Operator interface, including shopping, delivery, dining and travel – all of which allow for different types of automation.

When ChatGPT users activate an Operator, a small window will pop up showing the specific web browser the agent uses to perform the tasks, along with explanations of the specific actions the agent performs. While Operator is running, users can still control their screens because Operator uses its own dedicated browser.

OpenAI says the Operator is provided by a computer-based agent, or CUA, that incorporates the company’s vision capabilities. GPT-4o A model capable of reasoning from OpenAI’s more advanced models. CUA is trained to interact with the front-end of websites, meaning you don’t need to use developer-related APIs to use different services.

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

OpenAI says it’s working with companies like DoorDash, Instacart, Priceline, StubHub and Uber to ensure the Operator honors these businesses’ service contracts.

“The CUA model works before completing tasks with external side effects, such as submitting an order, sending an email, etc.,” OpenAI writes in a submission to TechCrunch. We aim to expand across a wide range of tasks.”

But OpenAI cautions that CUA isn’t 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 supervision for banking, CUA, and some tasks that the Operator can perform entirely on their own. Users will have to enter their credit card information, for example. OpenAI says Operator does not collect any data or take screenshots.

OpenAI says in its support materials: “On particularly sensitive websites such as e-mail, the Operator requires active user controls to ensure that users can immediately catch and eliminate potential errors that the model might make.

This limits the usefulness of the Operator, but also ensures that the agent doesn’t hallucinate and spend, say, your mortgage payment on accent chairs. Google took a similar approach Project Mariner An AI agent that doesn’t even fill in information like credit card numbers.

The agency will come

OpenAI has been quite slow to develop an AI agent compared to its competitors (see agents Rabbit, Googleand anthropic), may be related to security risks around the technology.

When an AI system can take action on the internet, it opens the door to more dangerous use cases from malicious actors. You can automate AI agents to orchestrate phishing scams or DDoS attacks, or snag tickets to a concert before anyone else. Especially for a tool as widely used as ChatGPT, it’s important for OpenAI to catch these bugs before they happen.

OpenAI thinks Operator is safe enough to release – at least as a trial.

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

Tasks gave ChatGPT users some familiar but necessary features to make ChatGPT as practical as Siri or Alexa. However, Operator demonstrates capabilities that previous generations of virtual assistants never could.

After ChatGPT, AI agents were billed as the next big thing in AI: a new technology that would change the way people use the internet and their computers. Instead of simply delivering and processing information, agents can theoretically take action and do anything.

As OpenAI publishes its first concrete treatment of agents, it will soon become clear how realistic this vision is.