The Hacker News
SOC leaders prepare to give AI agents greater autonomy
Security leaders are preparing to grant AI agents greater authority within Security Operations Centres, with a16z survey data showing comfort with autonomous low-risk actions rising from 36% today to 60% in 12 months. The piece, written by Zev Schonberg of Intezer, notes that AI-SOC initiatives emerged as the top priority for 2027 among CISOs at organisations with more than a thousand staff.
The driving pressure is the risk of AI-accelerated attacks outpacing response teams, alongside the reality that analyst time remains scarce and each alert requires cross-system evidence gathering.
Autonomy is framed as a spectrum rather than an all-or-nothing leap. The article traces a four-function automation model from Parasuraman, Sheridan and Wickens (information acquisition, information analysis, decision and action implementation) and maps it to SOC workflows: evidence collection, analysis and verdict, response recommendation, and response execution.
It emphasises that autonomy should be earned per action class, with the UW researchers’ five levels of human involvement (L1 to L5) used to classify where AI should operate. The piece cites a move from around L2 today toward L4 for low-risk actions within a year, while Gartner differentiates between human-in-the-loop and human-on-the-loop oversight.
Practical guidance includes a rubric for reversibility, blast radius and receipts, and a staged graduation framework (measured agreement, dry runs, complete evidence trails, rollback paths). The author also warns that granting autonomy increases potential damage if the agent is compromised, urging SOCs to treat AI agents like privileged identities and to enforce strict logging, scoping and auditing. The article is a contributed piece from Intezer.