OpenAI has introduced Operator, an innovative AI agent that can independently perform tasks online.
This development is a significant advancement in the company’s mission to improve AI agents, a technology that Sam Altman, CEO of OpenAI, has described as essential for the future by 2025.
Operator is a versatile AI tool capable of executing a variety of online tasks, including booking travel, making restaurant reservations, and shopping.
Initially, it will be available to US users subscribed to the $200 Pro plan, with plans to extend the research preview to users on Plus, Team, and Enterprise plans in the near future.
OpenAI also aims to broaden access to more regions, although users in Europe will need to wait a bit longer. The tool can be accessed at the dedicated site, operator.chatgpt.com, with intentions for future integration into all ChatGPT platforms.
Operator operates by controlling a specialized web browser, allowing it to perform tasks autonomously.
Users can observe the process, which mimics human interactions on websites, such as navigating menus, pressing buttons, and completing forms.
Image courtesy of OpenAI
Powered by OpenAI’s Computer-Using Agent (CUA) model, Operator combines the visual capabilities of GPT-4 with sophisticated reasoning skills. This model can manage a wide array of online tasks without the need for developer-facing APIs.
Its features extend to shopping, delivery services, dining, and travel bookings, with partnerships established with companies such as DoorDash, eBay, Instacart, Priceline, StubHub, and Uber to comply with their terms of service.
Before completing any action that has external consequences—like making purchases or sending emails—the CUA will request user confirmation, allowing users to review the actions beforehand.
Despite its advanced functionalities, OpenAI is aware that the CUA has limitations. Currently, Operator finds it challenging to manage complex or specialized tasks, such as creating intricate slideshows or overseeing detailed calendar systems.
Tasks that involve highly customized or non-standard web interfaces can also be problematic. Furthermore, activities like banking transactions necessitate user oversight to ensure they are conducted accurately.
While Operator can handle a broad range of tasks independently, OpenAI has implemented certain security protocols to mitigate the risk of misuse.
For instance, users must enter sensitive information such as credit card numbers manually, and for particularly sensitive activities like checking email, active supervision is required to prevent unnoticed errors.
There are also certain restrictions on the number of tasks Operator can execute. OpenAI has instituted dynamic rate limits for simultaneous tasks, along with an overall daily usage cap.
Additionally, some operations such as sending emails and deleting calendar events are restricted for security reasons, although these limitations may be revised in the future.
While Operator is OpenAI’s most advanced AI agent thus far, its development has been relatively slow compared to competitors like Rabbit, Google, and Anthropic. The advancement of such technologies raises serious safety concerns, especially regarding potential misuse.
To address these risks, OpenAI has incorporated safeguards into Operator, including monitoring systems that halt operations if suspicious actions are detected, along with ongoing updates to its security measures.
The launch of Operator signifies a landmark achievement in the evolution of AI agents, regarded as the next significant leap in artificial intelligence. Unlike conventional virtual assistants like Siri or Alexa, AI agents possess the ability to execute actions and tasks autonomously, reshaping how users interact with online services and the digital landscape.