ChatGPT for Google Sheets Plugin Exposes Serious Data Exfiltration Risks

• •ChatGPT for Google Sheets has serious vulnerabilities allowing data exfiltration.
• •Indirect prompt injection attacks can trigger unauthorized access to multiple workbooks.
• •Phishing overlays can deceive users into providing sensitive credentials.

A cybersecurity report from PromptArmor has revealed significant vulnerabilities in the 'ChatGPT for Google Sheets' plugin, which has over 185,000 downloads. The vulnerabilities stem from an indirect prompt injection attack that allows hackers to exfiltrate sensitive data across user accounts without authorization. Malicious scripts can be triggered by importing seemingly benign datasets, leading to the unauthorized access of financial models and other sensitive workbooks. Additionally, the attack can deploy phishing overlays that mimic official authentication pop-ups, tricking users into revealing their credentials. The report indicates that multiple workbooks can be compromised simultaneously, with one incident resulting in the exfiltration of 12 workbooks. OpenAI has been notified of these vulnerabilities but has not responded substantively. Security experts are advising users to review and manage permissions for AI extensions immediately.