ChatGPT flaw lets attackers pull Gmail data across accounts via a hidden channel

Summary

A flaw in OpenAI's ChatGPT allowed attackers to extract data from a victim's connected Gmail account by passing hidden instructions between separate user sessions. Researchers demonstrated how a victim's session could retrieve email data and relay it to an attacker-controlled session. OpenAI has since fixed the issue.

IFF Assessment

FOE

This vulnerability allowed attackers to exfiltrate sensitive data from connected services, posing a risk to user privacy and data security.

Defender Context

This incident highlights the risks associated with connecting AI models to sensitive enterprise applications and data. Defenders should scrutinize the security implications of AI integrations and monitor for similar cross-application data leakage vulnerabilities, especially in cloud-based services.

Read Full Story →