Exfiltration of personal information from ChatGPT via prompt injection

TL;DR

This paper demonstrates that ChatGPT 4 and 4o are vulnerable to prompt injection attacks that can exfiltrate personal data, especially with the new memory feature, raising privacy concerns.

Contribution

It reveals a novel security vulnerability in ChatGPT's architecture that enables data exfiltration through prompt injection without third-party tools.

Findings

Prompt injection can extract personal data from ChatGPT.

Memory feature increases vulnerability to data exfiltration.

All current ChatGPT users are potentially at risk.

Abstract

We report that ChatGPT 4 and 4o are susceptible to a prompt injection attack that allows an attacker to exfiltrate users' personal data. It is applicable without the use of any 3rd party tools and all users are currently affected. This vulnerability is exacerbated by the recent introduction of ChatGPT's memory feature, which allows an attacker to command ChatGPT to monitor the user for the desired personal data.