Ben Nassi, Bruce Schneier, and Oleg Brodt coined a new term and introduced a five-step kill chain model in their paper, "The Promptware Kill Chain: How Prompt Injections Gradually Evolved Into a Multi-Step Malware."

They mapped recent attacks to the new kill chain:

  1. Initial Access (prompt injection)
  2. Privilege Escalation (jailbreaking)
  3. Persistence (memory and retrieval poisoning)
  4. Lateral Movement (cross-system and cross-user propagation)
  5. Actions on Objective (ranging from data exfiltration to unauthorized transactions)

The authors aim to provide a shared vocabulary and methodology for threat modeling through this new model.

In my mind, it serves an even bigger role, elevating the conversation from a narrow "Can we block prompt injection?" to "How are we ensuring defense in depth?"

The Promptware Kill Chain: How Prompt Injections Gradually Evolved Into a Multi-Step Malware

Prompt injection is not SQL injection (it may be worse)