The Threat Isn’t the Frontier Model

This commentary urges security leaders to proactively develop and test AI-driven defensive agents and an AI control plane, prioritizing CTEM, breach-and-attack simulation, and security operations; it warns that advances in open-source models and quantization will soon lower the hardware barrier for adversaries, and recommends iterative, non-production testing, vendor partnership, and governance to maintain human oversight while automating repeatable security tasks.