Demystifying Security Risks of AI-Powered Applications on Pre-Trained Model Hubs · HackerLangs