Agentic Ransomware: From Human-Operated to AI-Operated Attacks

This report describes the rise of "agentic ransomware," where autonomous AI agents plan and execute ransomware campaigns end-to-end, and documents two concrete operations: FortiBleed (a large credential-harvesting campaign that used a 14-agent framework to build attack tooling and enabled human-operated intrusions across ~86,644 FortiGate devices) and JADEPUFFER (the first confirmed fully agentic LLM-driven extortion and destructive campaign exploiting Langflow CVE-2025-3248 and later deploying ENCFORGE to destroy AI/ML artifacts). It maps associated TTPs, provides IOCs for both campaigns, and recommends urgent defensive measures including prioritizing AI-adjacent asset hardening, patching, egress controls, secrets management, detection tuning for machine-speed behavior, and rehearsing incident response at automated speed.