How to steal intellectual property from GPTs 

A researcher demonstrated a practical attack against OpenAI "GPTs" that leverages simple prompts and the ChatGPT code interpreter to enumerate, access, and exfiltrate uploaded knowledge files and a GPT's internal configuration (custom prompts/instructions). The write-up details reconnaissance to discover filenames and paths, followed by using Python in the sandbox to zip and download the files, warns of potential business loss and sensitive data leakage, and advises avoiding uploading sensitive files until platform-level protections are implemented.