OpenAI has introduced ChatGPT Agent, an advanced AI assistant designed to autonomously perform complex tasks, marking a significant advancement in artificial intelligence capabilities. This new agent integrates features from previous models, such as Operator and Deep Research, enabling it to handle intricate workflows without continuous human oversight.
ChatGPT Agent can execute a variety of tasks, including planning events, analyzing data, and generating presentations. For instance, it can autonomously plan and purchase an outfit for a specific event, considering factors like weather and dress code. This functionality is achieved through a virtual computer equipped with tools that interact with the web and integrate with applications like Gmail and GitHub to gather relevant information for user prompts.
To ensure user control and safety, OpenAI has implemented several safeguards. The agent requests explicit permission before taking actions with real-world consequences, such as making purchases or sending emails. It also features a "Watch Mode" that pauses execution when the user becomes inactive, particularly during sensitive tasks like interacting with financial websites. Additionally, ChatGPT Agent is trained to identify and resist prompt injection attacks, where malicious instructions could manipulate its behavior.
Privacy controls are also a priority. Users can delete browsing data and log out of active sessions with a single click. During browser takeover mode, the agent does not collect or store sensitive information entered by the user, such as passwords, ensuring that inputs remain private.
ChatGPT Agent is currently available to subscribers of ChatGPT's Pro, Plus, and Team tiers. OpenAI plans to expand access to Enterprise and Education users in the coming weeks. As AI technology continues to evolve, ongoing vigilance and adaptation to emerging threats will be essential to ensure the safe and effective use of AI agents.