Threat Hunting

AI agent persistence attacks hide in prompt caches, tool registries, agent memory, and OAuth grants — surviving patches, restarts, and retraining. The five techniques and the hardening that stops them.