In a recent piece at MIT Sloan's Ideas Made to Matter (Church, 2025), Michael Siegel sets out three pillars for organizational defense against AI-enabled attacks: automated security hygiene, autonomous and deceptive defense systems, and augmented oversight with real-time intelligence. His central caution — that AI-powered cybersecurity tools alone do not suffice — is correctly placed. The multi-layered posture he describes is the right architecture for defending the modern enterprise against an adversary now equipped with generative models, deepfake pipelines, and automated reconnaissance at scale.
The three pillars define defense at the enterprise infrastructure layer. As organizations now deploy generative AI inside themselves — assistants, retrieval systems, copilots, and agents — Siegel's framework naturally extends to a new surface that did not exist as an attack vector five years ago. The asymmetry he names so cleanly in the piece — an attacker needs one entry point; a defender must guard all of them — applies just as well to this new surface. For a GenAI deployment, every prompt is a potential entry point.
Call this extension Pillar 00: real-time detection and adversarial testing at the input layer of every GenAI system the enterprise runs. The discipline isn't new in concept — Siegel himself points to artificial adversarial intelligence from MIT CSAIL, which mimics attackers to test defenses before real attacks arrive. Pillar 00 operationalizes that discipline at the prompt layer, in production, continuously.
The three pillars hold. Pillar 00 extends them to the GenAI surface. Build them alongside each other.