Operator, OpenAI’s AI agent capable of autonomously executing web-based tasks, arrives in EU

OpenAI has announced the deployment of Operator, its AI agent capable of accessing the web to perform the tasks you assign to it, in the European Union, Switzerland, Norway, Liechtenstein and Iceland, for Pro users.

This technology is still in the early research stages, but Operator is already available for subscribers to the Pro plan in the United States through this platform. OpenAI plans to test and improve it in a broad environment, learning from user feedback.

The developers intend to expand this feature to the Plus, Team, and Enterprise plans and integrate it into ChatGPT in the future.

📢 Operator Update

It’s now available to all Pro users in the EU, Switzerland, Norway, Liechtenstein & Iceland. https://t.co/aw0GIjHzCH

— OpenAI (@OpenAI) March 13, 2025

How Operator Works

Last week, OpenAI launched “Tasks,” a ChatGPT feature that allows scheduling tasks and reminders, marking a small step towards AI autonomy. Now, the developers have taken a decisive step towards true independence with Operator.

Operator is built on a new model called Computer-Using Agent (CUA), which interacts with graphical user interfaces (buttons, menus, texts, etc.) by combining the vision capabilities of GPT-4 with advanced reasoning through reinforcement learning.

According to OpenAI, if Operator encounters difficulties or makes errors while performing a task, it can use its reasoning abilities to self-correct. Additionally, “when it gets stuck and requires assistance, it simply hands control back to the user, ensuring a smooth and collaborative experience.”

Operator aims to streamline tasks for users and bring AI agents closer to businesses, allowing them to design innovative customer experiences that improve outcomes and increase conversions. OpenAI is collaborating with companies like DoorDash, Instacart, OpenTable, Priceline, StubHub, Thumbtack, Uber, among others, to ensure Operator meets real-world needs while adhering to established standards.

A research preview of Operator, an agent that can use its own browser to perform tasks for you. pic.twitter.com/wkBBDIlVqj

— OpenAI (@OpenAI) January 23, 2025

How to use it

Using this AI agent is straightforward. You just need to describe the task you want Operator to perform. After that, the agent will function autonomously, although in certain sensitive actions (logins, payments, CAPTCHA resolutions, email sends, etc.), it will request your intervention. You also have the authority to take control of the task at any time.

A tweet from Rowan Cheung highlights early access to ChatGPT Operator, showcasing its ability to autonomously perform web actions on behalf of users.

I got early access to ChatGPT Operator.

It’s OpenAI’s new AI agent that autonomously takes action across the web on your behalf.

The 9 most impressive use cases I’ve tried (videos sped up):

1. Ordering dinner ingredients based on a picture and a recipe pic.twitter.com/tdbApPELD4

— Rowan Cheung (@rowancheung) January 23, 2025

One notable feature is the ability to customize workflows in Operator by setting personalized instructions for all websites or specific sites. For instance, on the Booking website, you could define preferences for airlines. Additionally, Operator allows saving alerts and accessing them from the homepage, making it easier to handle repetitive tasks.

OpenAI states that Operator can perform multiple tasks simultaneously. This involves creating multiple conversations, each corresponding to a task, similar to opening several tabs in a traditional browser for different actions. However, the exact number of tasks that can be done concurrently has not been specified. When this limit is reached, users will receive a notification.

Regarding its limitations, “Operator cannot reliably or correctly perform certain complex tasks, such as creating detailed slide presentations, managing complex calendar systems, or interacting with highly customized or non-standard web interfaces.”

OpenAI will retain Operator data for up to 90 days

Because Operator captures screenshots to view the browser interface and perform actions, these images are recorded within the AI system. These screenshots, along with the chats with the agent and your browsing history, are stored in your account until you decide to delete them.

However, this does not mean that once you delete them, they will immediately vanish from OpenAI’s systems. “Users can delete browsing data or previous chats through the Operator settings page. When a chat is deleted, the screenshots taken during that chat are also removed. Deleted chats and associated screenshots will be erased from our systems within 90 days.”

The developer explains that they retain this data for a reasonable period to prevent misuse and ensure that their service is used appropriately. They also state that only a limited number of authorized OpenAI employees and trusted service providers have access to this data.

It is understandable to feel concerned or distrustful about the possibility of a company and its providers having access to screenshots that reveal sensitive information about your online activities, especially those related to transactions or personal or professional communications.

Photo: OpenAI