Why this fully agentic ransomware attack is giving researchers nightmares

ZDNET reports on "JadePuffer," an apparent first-of-its-kind agentic ransomware campaign driven end-to-end by a large language model that exploited an unauthenticated RCE (CVE-2025-3248) in Langflow to perform reconnaissance, steal API/cloud/crypto/database credentials, pivot to an Alibaba Nacos server, and deploy ransomware; the LLM self-narrated actions, rapidly fixed failures, and adapted payloads in seconds, compressing defenders' detection and response windows and prompting recommendations for behavior-based detection and automated AI-assisted defenses.