
The Path to AI Autonomy in the SOC: Survey Shows Growing CISO Comfort with Agentic Actions
A new a16z survey reveals that 36% of CISOs at large organizations are already comfortable allowing AI agents to act autonomously on low-risk security actions, with that figure expected to rise to 60% within 12 months. AI SOC initiatives rank as the top priority for 2027, driven by concerns that AI-accelerated attacks are outpacing human response capabilities. Researchers map autonomy across four workflow stages—evidence collection, analysis and verdict, response recommendation, and response execution—drawing from established aviation automation frameworks by Parasuraman, Sheridan, and Wickens. Five levels of agent autonomy are defined by the University of Washington, ranging from L1 Operator to L5 Observer, emphasizing that autonomy must be earned through measurable criteria rather than granted broadly. The article stresses treating AI SOC agents like privileged identities, implementing least agency principles from OWASP, and using reversibility, blast radius, and receipts rubrics to decide autonomy levels. Gartner’s distinction between human-in-the-loop and human-on-the-loop oversight is highlighted as essential for safe deployment. Intezer positions its forensic-level AI SOC approach as the mechanism that enables earned autonomy while mitigating risks such as prompt injection and compromised agents.
The Hacker NewsNov 20, 2026Translated from English