LecturaInvestigaciónRadarMarco de inversión
Iniciar sesión / Registrarse
← Volver a la lectura
NoticiasTema · AI智能体控制缺口Material1 min de lecturaEditorial de DisconfirmAI

Illumio Study Finds 70% of Organizations Delegate Decisions to AI Agents, But Only 49% Can Audit Them

Research reveals a significant 'control gap' in enterprise AI adoption, where autonomous agents make decisions faster than humans can verify their actions.

2026-10-30 04:20 GMT+8

A study by cybersecurity firm Illumio has identified a growing disconnect between the authority granted to AI agents in enterprises and the ability to monitor their actions. According to the report published on October 29, 2026, 70% of organizations are now delegating decisions that were previously made by humans to AI systems. However, only 49% of these organizations possess the capability to see and audit the specific actions taken by these AI agents.

This 'control gap' poses significant security risks. The study found that 76% of organizations experienced at least one security issue related to AI in the past year. Furthermore, 35% admitted that following AI recommendations would result in a security risk, while 26% reported that AI agents had connected to other systems without necessary approval.

Raghu Nandakumara, VP of Industry Strategy at Illumio, noted that the biggest risk may not be a system breaking rules, but one that was never given meaningful limits. He advised businesses to set strict boundaries on what AI agents can access and do, enforcing these limits within applications and infrastructure to contain unexpected behavior.

Fuentes

Iniciar sesión / Registrarse
LecturaInvestigaciónRadarMarco de inversión
techradar.com
TechRadar
↗
Tema

AI智能体控制缺口

网络安全公司 Illumio 发布的最新研究揭示了企业 AI 部署中的“控制缺口”:尽管 70% 的组织已将原本由人类做出的决策权移交给 AI,但仅有 49% 的组织能够完全审计 AI 智能体的行动。这一差距导致安全风险显著上升,因为 AI…

5 pasos1 piezasActualizado 2026-10-31 15:34

Cómo llegó hasta aquí

  1. 2026-10-21Illumio 发布《Licensed to Act: How AI Autonomy Is Outrunning the Limits Placed on It》报告,指出 70% 的组织将人类决策权移交给 AI,但仅 49% 能审计其行动。illumio.com ↗
  2. 2026-10-29TechRadar 报道 Illumio 研究,强调 AI 正在被赋予过多自主权来处理业务关键型安全决策,导致“控制缺口”。techradar.com ↗
  3. 2026-10-29TechTarget 分析指出,企业授予 AI 自主决策权的速度快于其安全保障能力,微软 CEO Satya Nadella 也呼吁遵循最小权限原则。techtarget.com ↗

Lo que aún está pendiente

  • OpenTelemetry 的生成式 AI 语义约定何时能发布稳定版本,以支持跨供应商的代理行为审计?
  • 监管机构是否会针对 AI 智能体的“不可解释性”和“难以中断性”出台具体的合规要求?
  • 企业在采用零信任架构时,如何平衡 AI 自主性与严格访问控制之间的效率损失?
Leer el tema completo →
Continuar investigando →