The latest innovation from OpenAI, known as Operator, is an AI agent capable of executing complex, multi-step tasks autonomously. This revolutionary tool was unveiled in a preview mode by the creators of ChatGPT, providing insights into its functionality and potential. The Operator, with the ability to surf the web, is adept at carrying out tasks such as calculating refunds for canceled orders, identifying clientele in a sales database based on specified criteria, purchasing groceries, and sending emails.
When deployed on a computer, Operator can perform a variety of tasks. These include downloading files, merging PDFs, analyzing spreadsheets, and exporting images. OpenAI aims to make 2025 the year of agentic AI. Last week, they launched Tasks for ChatGPT, a feature that allows users to automate regular prompts like sending a tech news digest or scheduling reminders. Coupled with the launch of Operator, which can autonomously handle more intricate tasks, OpenAI’s vision of making ChatGPT an essential tool utilizing its core product is becoming apparent.
Operator operates on a model known as a Computer-Using Agent (CUA) that fuses GPT-4o’s vision mode to view the user’s screen via screenshots and graphical user interfaces (GUIs) that allow Operator to interact with the screen through activities like clicking buttons, typing, and scrolling.
Operator: OpenAI’s Safety-First Approach
As a semi-autonomous AI agent, Operator raises significant safety concerns. However, OpenAI has assured that they have implemented multiple risk mitigation measures. This includes blocking Operator from carrying out harmful or illicit tasks and restricting access to blacklisted websites such as adult entertainment, gambling, and drug or gun retail sites.
OpenAI has also set up real-time automated safety checkers to review user interactions and ensure compliance with usage policies. These checkers have the authority to issue warnings or block prohibited activities. They have also developed automated detection and human review pipelines to identify and prevent prohibited usage in critical policy areas such as child safety and deceptive activities.
Given that Operator can make costly errors without human oversight, the model will seek user confirmation before finalizing actions like placing an order or sending an email. This gives the user an opportunity to review the model’s work before it is finalized. For added safety, Operator is currently prohibited from “high-risk tasks” such as banking transactions.
Availability of Operator
OpenAI has introduced a new premium tier subscription known as ChatGPT Pro. The preview mode of Operator is currently available only to U.S. users who subscribe to the Pro plan at $200 per month. However, OpenAI plans to gradually expand availability to Plus, Team, and Enterprise users in the future.
