PromptArmor: Ramp's Sheets AI can be prompt-injected to exfiltrate financials

Security researchers at PromptArmor demonstrated that Ramp's Sheets AI can be prompt-injected to exfiltrate financial data, intensifying enterprise concerns about agentic LLM deployments. The writeup hit 124 HN points; Mandiant separately warned a malicious npm dependency tied to an AI-assisted commit targeted crypto wallets.