OpenAI launches Operator, an AI agent that can perform tasks on the web

Surprisingly, the OpenAI released the first version of an AI agent that has everything it needs to be a personal assistant for many people. Operator can make purchases, read PDF files and more tasks, with just a text command. See details and who can use it.

What is agentic artificial intelligence?

A agentic artificial intelligence refers to AI systems designed to act as autonomous agents, capable of making decisions, planning actions, and performing tasks independently. These agents are guided by clear goals and use information from the environment to act proactively and adaptively, similar to a human or robotic agent.

Key features of agentic AI include autonomy, environmental awareness, algorithmic decision-making, and continuous learning to adapt to new conditions. Practical examples include autonomous robots, intelligent virtual assistants, game characters, and industrial automation systems, all of which operate independently to achieve specific goals.

The central difference between agentic AI and other approaches is its ability to decide how and when to perform tasks in dynamic environments, going beyond predefined roles. However, this flexibility requires careful attention to safety, ethics, and control issues to ensure that agents act in a manner that is aligned with human interests.

Everyday use of Operator

During a half hour online event, we were introduced to what the Operator can do. When you type a command to the AI ​​agent, you will need to select the GPT for action, which can be defined as a “specialist” of the task. For example, if you need to shop online, you will need to select the InstaCart, as in the image above.

Users can also request that the Operator Check inventory before going to a physical store by simply entering the address, item name and size. The response will be sent in a few seconds and you will also receive route tips to get to the location in the shortest time possible.

One of the most interesting tasks is purchasing tickets. Simply type the text command stating the event and type of ticket you would like and wait to be redirected to the official sales website. A screen with all available tickets will be displayed and the user only needs to choose the sector they want to be directed to the final purchase screen.

The tool can also understand photos. A practical example of this is that you can send a shopping list and ask for the items to be purchased, all you need to do is take care of the payment.

Basically, all the tasks that the GPTs are implemented at Operator da OpenAI, but the very Sam altman highlighted that agentic artificial intelligence may present errors, as it is still under development. The showmetech has a complete article with the best GPTs To find out how and in what situations you can use each of them, be sure to check it out.

Security

A OpenAI is aware of how important sensitive data is, so when the Operator need to do some identity confirmation, it will be necessary for a human to do this work, such as, for example, logging into an account or even entering credit card details. In order for the data during use, the AI ​​agent of the Chat GPT has special modes, which are:

• Takeover Mode: o Operator prompts the user to take control by entering sensitive information into the browser, such as login credentials or payment information. When in Takeover mode, the Operator does not collect or capture user-inputted information. 
• Task limitations: o Operator is trained to refuse certain sensitive tasks, such as banking transactions, or those that require high-risk decisions, such as making a decision about a job application.
• Observation mode: On particularly sensitive sites, such as email or financial services, the Operator requires close supervision of their actions, allowing users to directly identify any potential errors.

A OpenAI also put in place a user confirmation tool, which requires the human to give the OK to perform a more rigid action, such as closing a purchase or sending an email; it is a two-factor authentication, but implemented in the AI ​​agent. OpenAI.

In the settings, users will be able to disable their usage from being used to train the AI ​​agent. Browsing data can also be completely erased at any time. Finally, a dedicated “monitor model” watches for suspicious behavior and can pause the task if something seems strange.

Despite the details, the company run by Sam altman know that malicious people can use the Operator, so expect the tools to be a way to prevent potential scams.

OpenAI on its AI agent.

We know that bad actors may try to misuse this technology. That’s why we designed Operator to refuse harmful requests and block unauthorised content. Our moderation systems can issue warnings or even revoke access for repeated violations, and we’ve built in additional review processes to detect and address misuse. While Operator is designed with these safeguards in mind, no system is perfect and this is still a research preview; we’re committed to continuous improvement through real-world feedback and rigorous testing.

Price and availability

At this first moment, the Operator will only be available to US subscribers of ChatGPT Pro plan, which costs US$200 per month (R$1.187,86). The company explains that the new feature may be released in the future for more countries and subscribers of the ChatGPT Plus plan, but there is still no set date for this to happen.

For now, despite sounding innovative, the Operator continues to be developed and a final version should be released, just as happened with the Sora, the AI ​​that creates videos. Anyway, what do you think of the new product? Would you use it in your daily life? Tell us Comment!

See also other features

Text revised by Alexandre Marques in 23 / 01 / 2025.

With information: OpenAI