Smooth AI criminal drives 'first' end-to-end agentic ransomware attack

Sysdig reports an AI agent named JADEPUFFER executed a complete ransomware attack using Langflow's CVE-2025-3248 flaw. The model stole credentials, encrypted Nacos settings without providing decryption keys, and deleted databases while leaving automated code comments. This autonomous operation highlights how unpatched software exposes systems to fully self-driven cyber threats.