OpenAI has launched a new AI agent in ChatGPT that can perform a wide range of tasks on behalf of users. The company says the agent can automatically navigate a calendar, create editable presentations and slideshows, and run code.
OpenAI has launched a new AI agent in ChatGPT that can perform a wide range of tasks on behalf of users. The company says the agent can automatically navigate a calendar, create editable presentations and slideshows, and run code.
The tool, called ChatGPT agent, combines several features from OpenAI’s previous agent tools, including the ability for an operator to click through websites, as well as Deep Research. Users will be able to interact with the agent simply by typing queries into ChatGPT in natural language, TechCrunch reports .
The ChatGPT agent was rolled out yesterday to OpenAI Pro, Plus, and Team subscribers. To activate the tool, users can select “agent mode” from the ChatGPT tool drop-down menu.
OpenAI claims that this ChatGPT agent is much more powerful than previous versions.
The new agent can access ChatGPT connectors, which allow users to connect to applications such as Gmail and GitHub so that the agent can find relevant information for user queries. The ChatGPT agent also has access to the terminal and can use APIs to access certain applications.
The model underlying the ChatGPT agent offers state-of-the-art performance across multiple benchmarks, OpenAI claims.
The company claims that the ChatGPT agent model scores 41.6% on Humanity's Last Exam (pass@1), a complex test consisting of thousands of questions across over a hundred subjects. This is about double the score of OpenAI's o3 and o4-mini on this test.
On FrontierMath, one of the most difficult math benchmarks known, OpenAI claims that the ChatGPT agent scores 27.4% when it has access to tools like a terminal to execute code. A recent result was given to o4-mini, which scored just 6.3%.
OpenAI notes that it developed the ChatGPT agent with security in mind, mainly because the product has some new capabilities that could make it more dangerous in the hands of attackers.
In a security report on the ChatGPT agent, OpenAI notes that it has identified the model as “highly effective” in the field of biological and chemical weapons. The company adds that it has no direct evidence of this, but decided to take a precautionary approach and activate new safeguards to mitigate such risks.
New safeguards for the ChatGPT agent include a monitor that runs in real time as users interact with the product. OpenAI says it runs a classifier on each query entered into the ChatGPT agent, determining whether the query is related to biology. If so, OpenAI runs the ChatGPT agent’s response through a second monitor that determines whether the content could be used to trigger a biological threat.
OpenAI also says it has disabled ChatGPT’s memory feature for this agent to prevent abuse. In other parts of ChatGPT, OpenAI’s memory feature allows the chatbot to access information from users’ previous chats. However, OpenAI says that attackers could exploit this feature in the ChatGPT agent to extract sensitive data through rapid deployment attacks. But the company may return to adding this feature in the future.
