OpenAI's ChatGPT Agent Launches with 'High' Biorisk Classification, Automating Complex Tasks

OpenAI has officially launched ChatGPT Agent, a new artificial intelligence tool designed to perform complex tasks autonomously on behalf of users. The release, which began rolling out to select subscribers on July 17, 2025, marks a significant shift from traditional chatbots to AI capable of taking direct action. Tech enthusiast Jacklyn Dallas, who gained early access, lauded the update on social media, stating, "AI that can actually do things on your behalf."

The ChatGPT Agent integrates the functionalities of OpenAI’s previous tools, Operator and Deep Research, allowing it to navigate virtual browsers, run code, and interact with third-party applications. Users can instruct the agent to manage calendars, book reservations, analyze financial data, build Excel models, and generate presentations. Dallas highlighted its practical utility, noting it "takes your goal and then it figures out all the steps to make it happen. And then it executes them!"

Initially, ChatGPT Agent is available to Pro, Plus, and Team subscribers, with Enterprise and Education users gaining access in the coming weeks. Pro users are allotted 400 agent prompts monthly, while Plus and Team tiers receive 40. Users can activate the feature by selecting "agent mode" within ChatGPT, and the system is designed to pause for user confirmation before executing significant actions.

Notably, OpenAI has classified ChatGPT Agent as having a "high" capability for biorisk, meaning it could potentially assist "novice" actors in developing biological or chemical threats. To mitigate these concerns, the company has implemented robust safeguards, including refusing prompts related to biological threats, real-time content monitoring, and disabling the memory feature to prevent misuse. OpenAI emphasizes user control, allowing interruptions and takeovers during tasks.

This launch represents OpenAI's boldest move into agentic AI, aiming to provide a "genius on your team" that can automate daily and professional workflows. The company reports the agent achieves state-of-the-art performance on several benchmarks, including a 41.6% score on Humanity’s Last Exam and significant improvements in data science and spreadsheet tasks. The development signals a broader industry trend towards more autonomous AI systems.