Four ways AI has fundamentally changed the threat landscape in 2026

Sysdig Threat Research documents a new class of autonomous, agentic AI attacks that plan, execute, and adapt without human-in-the-loop — exemplified by JADEPUFFER (agentic ransomware) and other incidents that rapidly exploit disclosed flaws, steal credentials and keys from AI infrastructure (LLMjacking), escape containers to access Kubernetes secrets, and exfiltrate data; the report emphasizes the compressed exploitation timeline (minutes), distinct behavioral signals of LLM-generated payloads, and recommends runtime visibility, identity controls, and an AI bill of materials to mitigate risk.