OpenAI has launched a research preview of Operator, its first general-purpose AI agent that can control a web browser to perform tasks independently.
- Where to find it: U.S.-based users on ChatGPT’s $200 Pro subscription plan can try Operator first via operator.chatgpt.com.
- Coming soon: OpenAI plans to expand access to users on its Plus, Team, and Enterprise tiers. CEO Sam Altman noted during a Thursday livestream that it will roll out to other countries, though Europe may face delays.
How it works
Operator uses a built-in web browser to automate tasks like making restaurant reservations, shopping, and booking travel.
- Inside Operator: Users see a dedicated browser window where the AI interacts with websites — clicking buttons, navigating menus, and filling out forms, much like a human.
- Human oversight: Operator asks for user confirmation before finalizing tasks with external consequences, like placing orders or sending emails.
- Safety first: Sensitive actions, like entering credit card details, require users to step in. OpenAI says Operator doesn’t collect or screenshot any data.
Zoom out
AI agents like Operator are a leap beyond Siri or Alexa, aiming to execute actions rather than just provide information.
- The tech behind it: Operator is powered by OpenAI’s computer-using agent (CUA), which combines vision and reasoning from advanced models. It interacts with websites directly — no developer APIs required.
- Corporate partnerships: OpenAI is collaborating with companies like DoorDash, Instacart, and Uber to ensure compliance with terms of service agreements.
Yes, but: Operator isn’t perfect yet.
- OpenAI warns the CUA may not perform reliably in all scenarios. For sensitive sites like email, Operator requires active user supervision to prevent mistakes.
- The risks of AI agents taking actions on the web — from phishing scams to misused automation — have slowed development across the industry.
Source: Techcrunch
