Defense in Depth for AI Agents: From Injection to Autonomy
No single defense protects autonomous agents. A synthesis of six recent papers tracing the layered threat landscape — from prompt injection through behavioral contamination to goal drift — and arguing that only their combination offers meaningful protection.
Isman Fairburn