The vulnerability stems from the Code Interpreter plugin in gpt_academic, which executes user-provided prompts to generate code. Due to the lack of proper sandboxing, an attacker can manipulate the prompt to include malicious code, leading to remote code execution on the server. This allows the attacker to gain full control over the backend server, posing a significant security risk.

An attacker can exploit this vulnerability by selecting the Code Interpreter plugin and uploading a file. They then provide a specially crafted prompt that includes malicious code. When the plugin processes this prompt, it executes the malicious code, allowing the attacker to run arbitrary commands on the server. For example, the attacker can use the prompt to execute the env command or create a file on the server.

Users and administrators of the gpt_academic application using the Code Interpreter plugin are affected. This includes any deployment of the application running version 3.83 or earlier.