The Promptware Kill Chain

This essay introduces the "promptware kill chain," a seven-stage framework for understanding multistage attacks against LLM-based systems, covering initial access, jailbreaking, reconnaissance, persistence, C2, lateral movement, and actions on objectives. It argues that prompt injection is only the first step in broader malware-like operations and uses research demonstrations (e.g., malicious prompts in calendar invites and emails) to show persistence and propagation, urging defenders to assume initial access and disrupt subsequent stages through layered controls.