OpenAI CEO Sam Altman warned users to exercise caution when using the new ChatGPT agent. This agent is the company’s first system capable of performing multi-step tasks independently. It features an advanced set of functions, breaks down requests into separate stages, uses external tools, and operates without human intervention.
Altman emphasizes that the agent should not be used for important matters or processing personal information. Despite numerous safeguards, the system remains experimental. He noted that if the agent is granted broad permissions, such as access to email, there is a risk of unauthorized actions or data leaks through malicious messages.
Researchers have repeatedly demonstrated that AI agents can be vulnerable to manipulation through simple prompts. This can lead to unwanted disclosure of private information or execution of incorrect actions. Altman recommends limiting the agent’s access to only the permissions necessary for a specific task to reduce security and privacy risks for users.
OpenAI plans to enhance the agent’s protection based on user feedback and real-world situations. The company emphasizes that responsibility for using the agent and any potential consequences lies with the user. Before using the new feature, users are advised to carefully review its capabilities and risks.
