Operator, OpenAI's agent that can perform multi-step tasks autonomously, has arrived.
The ChatGPT maker introduced a preview mode of Operator on Thursday, detailing how it works and what it's capable of. Operator can browse the web, performing tasks like calculating refunds from a canceled order and finding customers with specific criteria in an internal sales database. It can also buy groceries and send emails.
On a computer, Operator can perform various tasks, like downloading files, combining PDFs, analyzing spreadsheets, and exporting images.
OpenAI is delivering on its promise of making 2025 the year of agentic AI. Last week, the company launched Tasks for ChatGPT, which lets users automate future prompts like sending a daily brief of tech news or scheduling reminders. While many of these tasks are already possible through basic tools like Google Alerts and calendars, it's an early example of AI bots doing the legwork for the user. Combined with the release of Operator and its ability to autonomously take on more complex tasks, you can start to see OpenAI's vision for making ChatGPT an indispensable tool leveraging its core product.
The model underpinning Operator is a Computer-Using Agent (CUA) that combines GPT-4o's vision mode to "see" what's on the user's screen through screenshots with graphical user interfaces (GUIs) that enable Operator to interact with the screen (clicking buttons, typing, scrolling, etc.).
OpenAI's safety approach with Operator
Obviously, safety is a big concern for a semi-autonomous AI agent like Operator. OpenAI says it has taken risks into account in a few different ways. Operator mitigates misuse by blocking harmful or illegal tasks, and can't access blacklisted sites like gambling and adult entertainment sites and drug or gun retailers.
And OpenAI is looking over your shoulder as you use Operator. The announcement says that "user interactions are reviewed in real-time by automated safety checkers that are designed to ensure compliance with Usage Policies and have the ability to issue warnings or blocks for prohibited activities," and that the company has developed "automated detection and human review pipelines to identify prohibited usage in priority policy areas, including child safety and deceptive activities."
Since Operator can make costly mistakes without human supervision, the model will ask for confirmation "before submitting an order, sending an email, etc., so that the user can double-check the model’s work before it becomes permanent." Operator is also currently limited from "higher-risk tasks like banking transactions."
Operator availability
Now is where we start to see OpenAI's new premium tier subscription, ChatGPT Pro. Operator in preview mode is only available in the U.S. to those who pay $200 a month as Pro users. But over time, OpenAI expects to expand availability to Plus, Team, and Enterprise users.
