AEGLIX News
A survey of chief information security officers at organizations with more than 1,000 employees reveals that 36% are currently comfortable allowing AI agents to act autonomously on low-risk security actions, according to research from venture capital firm a16z. The same study found that 60% of respondents expect to reach that comfort level within 12 months, signaling a rapid shift toward autonomy in security operations centers. However, only 1% of CISOs said they are comfortable with AI agents acting autonomously on most actions today, and just 7% anticipate that level of trust within a year.
The survey highlights a tension between growing adoption and persistent concerns. More than half of CISOs, 55%, identified AI-accelerated attacks outpacing their team's ability to respond as their most common worry, while 29% cited prompt injection or data leakage through AI applications. Despite these fears, deployment is already underway: 55% of respondents reported running MCP servers or other agent-to-application integrations in production, and 34% said they operate autonomous agents with tool access or write permissions. A majority, 52%, want AI agents to act only with human approval, reflecting a preference for supervised autonomy.
The findings align with a long-standing automation framework proposed by Parasuraman, Sheridan and Wickens in a widely cited 2000 paper, which defines four broad functions of automation: information acquisition, information analysis, decision and action selection, and action implementation. The current trajectory suggests organizations are moving from analysis toward decision and action selection, with limited but expanding implementation authority. The data indicates a pragmatic approach where low-risk, repetitive tasks are the first to be delegated, while high-impact decisions remain under human control.
The article recommends treating the AI SOC agent like a new privileged identity by giving it its own identity, scoping permissions, and logging every tool call, query and action. This approach mirrors identity governance practices applied to human administrators and service accounts, extending them to autonomous software agents operating within the security environment. By enforcing least-privilege access and maintaining an immutable audit trail of agent activity, organizations can establish accountability and reduce the risk of unauthorized or unintended actions as autonomy expands.
Report prepared 2026-11-20 · Based on reporting from The Hacker News. Source publication date above; subsequent developments may not be reflected.