Poisoned documents can turn reasoning-based AI guardrails into DoS weapons by leveraging security systems themselves as resource sinks—a new attack vector with concentration risks in shared governance infrastructure.
The Heretic tool can remove security filters from open-source AI models in minutes—a structural control risk that undermines existing compliance frameworks for locally deployed models.
