OpenAI, a research organization known for its AI models, has introduced the ChatGPT Agent, a system designed to perform complex tasks autonomously. Available to Pro, Plus, and Team users, the agent can analyze data, browse the web, execute commands, and generate reports or presentations. It combines browsing, coding, and app integration through tools like visual and text browsers, a terminal, and direct API access. The agent uses its virtual computer to maintain context across tasks, shifting between reasoning and action as needed. “ChatGPT is trained to explicitly ask for your permission before taking actions with real-world consequences, like making a purchase,” OpenAI confirmed. Users remain in control, with options to pause, intervene, or supervise tasks, while the model seeks clarification when required.
The agent integrates features from OpenAI’s earlier Operator and Deep Research tools, enhancing both task execution and information gathering. Its performance has set new benchmarks in evaluations like Humanity’s Last Exam and DSBench. OpenAI also addressed safety concerns, implementing measures against prompt injection attacks and model misuse. With privacy controls, secure browser sessions, and explicit action confirmation, the system emphasizes user oversight. Though some capabilities, like slideshow generation, remain in beta, OpenAI plans continuous updates. “This launch is just the beginning,” the company stated, signaling ongoing development.