OpenAI has unveiled a new ChatGPT feature called Agent Mode, allowing artificial intelligence to operate within a virtual computer and perform complex tasks autonomously, while always under human supervision.

Nicknamed “Skynet mode” on social media in reference to Terminator, this tool represents a major step in AI-assisted automation for everyday tasks. 

Since mid-July, subscribers to paid plans can enable this feature, creating a virtual environment where the model can carry out multiple chained actions: browsing websites, editing documents, filling out forms, creating presentations, and even executing and debugging code snippets.

All of this occurs in an isolated session, protected by multiple layers of security. 

To activate Agent Mode, users simply type /agent in a ChatGPT conversation or select it from the tools menu. Once started, the model launches a “virtual computer” that responds in real time. For critical actions, such as sending emails or financial transactions, explicit user confirmation is required before proceeding. 

Tasks the new Agent Mode can perform include generating complete presentations from meeting calendars, searching the web and transferring information to documents, filling spreadsheets, editing files, and automating multi-step processes that previously required several windows and manual actions.

However, the agent cannot send emails independently, make payments, or control devices outside the virtual environment, maintaining human oversight at all times. 

OpenAI and Skynet mode

OpenAI emphasized that the system can detect and block malicious requests, including dangerous prompt injection attempts. Furthermore, if the user switches tabs or becomes inactive during a critical task, the session automatically stops, ensuring operations remain under constant supervision. 

The launch of Agent Mode has reignited debate about the limits of human control over artificial intelligence.

Although the nickname “Skynet” circulates widely on social media, OpenAI stresses that this is a collaborative and secure tool. “ChatGPT only acts when prompted, and every step can be reviewed, paused, or reversed,” the official announcement stated. 

Currently, this feature is available to Pro, Plus, and Team plan users with varying monthly interaction limits. OpenAI plans to expand it soon to educational and enterprise accounts. Early adopters note that this mode represents the closest step yet toward intelligent automation of daily tasks, while maintaining human supervision and control.