The AI Cybersecurity Arms Race Is On
CIO, Monday, September 7th, 2026
Agentic AI incidents signal an arms race in which defenders must rely on AI to protect against AI-powered attacks.
Rising cyberattacks and recent incidents in which AI agents broke into systems, including an intrusion at Hugging Face, suggest agentic attacks will surge.
The author argues attackers hold the advantage because guardrails on US frontier models block defensive use; Hugging Face turned to a Chinese open weight model to analyze attack logs.
Agent swarms, exposed MCP servers, deepfake voice scams and an agent that created fake identities to push malicious code widen the threat.
The piece calls for giving developers and security vendors less restricted access to frontier models, citing lessons from 1990s US encryption export limits.