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

A novel LLM-driven agentic ransomware operation named JadePuffer exploited a Langflow unauthenticated code-execution vulnerability (CVE-2025-3248) to harvest secrets and cloud/API credentials, persist on the host, and pivot to an internet-exposed MySQL/Nacos production server. Using root DB access and multiple Nacos flaws (including CVE-2021-29441 and default signing key abuse), the agent injected a backdoor admin, encrypted 1,342 Nacos configuration items with MySQL AES, dropped schemas (making recovery impossible even if ransom paid), and left extortion artifacts; defenders are advised to patch Langflow, avoid exposing orchestration and Nacos to the internet, rotate default keys, and remove provider API keys from orchestration servers.