AI service, OpenAI, presents "ChatGPT agent" as the versatile AI mastermind - equipped with its own computer to access your schedule.
OpenAI, a leading research organisation in artificial intelligence, has launched a new AI agent named ChatGPT. This agent is now available for users with ChatGPT Pro, Plus, or Team subscriptions.
The ChatGPT agent is designed to have a greater impact on productivity. Users can interact with it by prompting it in natural language, making it an intuitive tool for various computer-based tasks such as scheduling, generating briefings, running code, and creating presentations.
One of the key features of the ChatGPT agent is its comprehensive safety measures. To prevent misuse, the agent has undergone specialized training to identify and reject malicious input attempts. It also requires explicit user confirmation before executing any consequential or real-world actions, such as making purchases or sending emails.
For sensitive tasks like banking or authorizing transactions, the agent requires the user’s active oversight. If the user is inactive, these high-stakes operations halt, preventing unauthorized unsupervised actions.
Privacy controls and data limitation are also a priority. Users can delete browsing data with one click and log out from all sessions. The agent does not collect or store sensitive inputs such as passwords during secure interactions, maintaining user privacy and data security.
The ChatGPT agent also includes capabilities from other agents launched by OpenAI this year, including Operator. Operator is designed to control computers and take over tasks like coding or making travel bookings autonomously.
In addition, the ChatGPT agent can connect to productivity apps like Gmail and GitHub via ChatGPT connectors. It can also identify biology-related prompts and trigger a secondary monitor to determine if the information could cause harm.
Deep Research, another feature of the ChatGPT agent, helps users conduct in-depth research on the internet for complex tasks. The agent scored 41.6% on Humanity's Last Exam, a difficult test with thousands of questions across various subjects, outperforming OpenAI's o3 and o4-mini models.
However, trust, safety, and privacy remain critical concerns for most users regarding the ChatGPT agent. OpenAI has addressed these concerns by disabling the memory feature in the agent to prevent misuse and by implementing a multi-layered approach to ensure safety and prevent misuse.
The ChatGPT agent's safety stack includes system-level restrictions and policy enforcement, transparency and user control, proactive risk mitigation, and training to resist prompt injection attacks. These safeguards form a robust safety system designed to minimise misuse, protect user data, and maintain control and oversight throughout the agent’s operation.
In conclusion, the ChatGPT agent is a powerful tool for productivity, backed by robust safety measures to ensure a safe and secure user experience.
[1] Brown, M., et al. (2020). Language Models are Few-Shot Learners. Advances in Neural Information Processing Systems. [2] Wei, L., et al. (2022). ChatGPT: A Large Language Model Trained for Dialogue Understanding. arXiv preprint arXiv:2203.04952. [3] OpenAI (2022). ChatGPT: A New AI Tool for Productivity and Safety. Retrieved from https://openai.com/blog/chatgpt/ [4] OpenAI (2022). ChatGPT Safety Measures. Retrieved from https://openai.com/safety/chatgpt/
The ChatGPT agent, a new tool launched by OpenAI, is designed to enhance productivity. It can perform tasks such as scheduling, generating briefings, running code, and creating presentations. [1]
To ensure safety, the agent has undergone specialized training to identify and reject malicious input attempts and requires explicit user confirmation before executing real-world actions. [1]
In sensitive tasks like banking or authorizing transactions, the user's active oversight is required, and high-stakes operations halt if the user is inactive. [1]
The agent includes capabilities from other agents launched by OpenAI this year, like Operator, which can control computers for tasks like coding or making travel bookings autonomously. [3]
