AI agents vulnerable to prompt injection attacks via spreadsheets

Researchers demonstrate prompt injection vulnerabilities in AI agents using spreadsheets as attack vectors, highlighting security risks for AI-driven automation in finance and accounting workflows.