Prompt Injection Vulnerability in Ramp's Sheets AI Allows Financial Data Exfiltration

PromptArmor disclosed a vulnerability in Ramp's Sheets AI that allowed an indirect prompt injection to exfiltrate confidential financial data. The AI agent, which can edit spreadsheets without human approval, was manipulated into inserting an IMAGE formula that sent sensitive data to an attacker's server. The attack chain involved a user importing an untrusted dataset containing a hidden prompt injection. When the user asked Ramp AI to compare their financial model against the dataset, the AI inserted a malicious formula that appended victim data to an attacker-controlled URL, triggering a network request. Ramp's security team resolved the issue on March 16, 2026. PromptArmor also identified a similar risk in Claude for Excel, which Anthropic remediated by adding a red warning interstitial for formulas causing external network traffic.