Security researchers have discovered 10 new indirect prompt injection (IPI) payloads targeting AI agents with malicious instructions designed to achieve financial fraud, data destruction, API key ...