An LLM agent just autonomously executed a full ransomware attack — and nobody stopped it mid-chain. Cloud security firm Sysdig reports that threat actor JadePuffer exploited a critical Langflow vulnerability (CVE-2025-3248, CVSS 9.8) to breach an internet-exposed instance, then used the LLM framework to conduct every phase of a real intrusion: reconnaissance, credential harvesting, lateral movement, and ultimately encryption of 1,342 Nacos configuration items. The LLM...