OpenAI has launched ChatGPT Agent, an AI agent that combines ChatGPT, deep research, and Operator. During the live stream, the team spoke about how ChatGPT Operator and deep research are complementary. How, you may ask? ChatGPT Operator, the computer use agent, while having the capability to interact with websites and perform tasks on the user's behalf, isn't as effective as reading long articles. On the other hand, while ChatGPT's deep research is good for conducting online research, it can interact with the web to complete tasks.
The OpenAI team found out that users were asking the ChatGPT deep research agent to find information and also perform tasks. Hence, the introduction of ChatGPT Agent, which combines the ChatGPT deep research agent that can conduct in-depth web research and ChatGPT Operator, the computer use agent that can perform tasks.
ChatGPT agent with a suite of tools that allows it to conduct research and perform tasks. ChatGPT Agent combines a visual browser that interacts with the web through a graphical user interface, a text-based browser for simpler reasoning-based web queries, a terminal, and direct API access. The agent uses ChatGPT connectors, allowing users to connect apps like Gmail and GitHub so ChatGPT can find information relevant to their prompts and use them in its responses.
![ChatGPT Agent [Findora]](https://aitoolsclub.com/content/images/2025/07/Untitled-design--28--1.gif)
OpenAI also spoke about the prompt injection, which is a cybersecurity exploit that tricks large language models (LLMs) into performing an unintended action. For example, AI agents are supposed to be autonomous and perform tasks to help users, but they can be tricked by malicious websites.
Suppose you want the ChatGPT AI agent to make an online purchase on your behalf, but it can be tricked by a malicious website that asks it to enter credit card details. OpenAI says it is trying to tackle these, but it recommends users to use the handover feature, where, when it is time to add sensitive details, the ChatGPT Agent can hand over control back to the users.
Source: OpenAI
