Vulnerability Discovered That Lets AI Insert Formulas into Spreadsheets and Send Sensitive Data Externally

PromptArmor reported that Ramp Labs' Sheets AI could be tricked by indirect prompt injection into auto-inserting a formula that sends data externally.

The flaw could leak confidential financial data from spreadsheets to an attacker-controlled server without user approval.

Ramp fixed the issue on March 16, 2026.

The case highlights a major security risk for business AI tools that can create or modify spreadsheet formulas.