Security · 1h ago
IBM Engineer Details 7-Stage Promptware Kill Chain for AI Attacks
IBM distinguished engineer Jeff Crume outlines a new threat model called Promptware, where malicious prompts drive AI agents to execute attacks. The kill chain includes initial access via direct or indirect injection, privilege escalation, reconnaissance, persistence, command & control, lateral movement, and action on objectives. Crume warns that prompt injection is architecturally unfixable and recommends zero-trust defenses.
Meridian48 take
The Promptware kill chain is a useful framework, but its claim that prompt injection is 'architecturally unfixable' may overstate the case—defenses like input sanitization and strict API controls are emerging.
prompt-injectionai-security