LesenForschungRadarAnlageframework
Anmelden / Registrieren
← Zurück zur Lektüre
NewsThema · 美陆军AI网络代理Wesentlich2 Min. LesezeitDisconfirmAI Redaktion

US Army reportedly developing autonomous AI cyberattack agents, experts warn of loss of control

According to The Intercept, the US Army is developing AI weapons capable of autonomously seeking and destroying targets, with experts warning of catastrophic risks due to lack of human judgment.

2026-10-20 21:30 GMT+8

According to a report by The Intercept on October 19, 2026, the U.S. Army Cyber Command (ARCYBER) is developing a program codenamed "ECHOTRIBBLE," designed to deploy autonomous AI agents capable of navigating enemy digital infrastructures, identifying targets, and "neutralizing" them. These agents are purportedly designed to function like computer worms, moving between devices and operating systems, self-replicating, and self-destructing after completing their mission to minimize traces.

The project is currently in the early prototyping and industry assessment phase. Leaked documents indicate that ARCYBER held a virtual assessment event in early October 2026, narrowing down five candidate solutions from 65 technical whitepapers for further review. The military seeks agents capable of "Self-Directed Behavior," operating with minimal human oversight to achieve the desired level of autonomy.

Cybersecurity experts and AI researchers have raised serious concerns. David Krueger, a machine learning professor at the University of Montreal, stated that given the known failure modes of current AI paradigms, granting autonomous systems a mandate to "neutralize threats" is incredibly dangerous. Hyrum Anderson, CEO of Abundant, warned that autonomous agents could inadvertently damage civilian infrastructure, such as U.S. water utilities, if adversaries stage command and control facilities within critical nodes. William Marcellino, a senior behavioral scientist at the RAND Corporation, emphasized that current public AI technology lacks the world knowledge and moral judgment possessed by human soldiers, who know "what they probably shouldn't do."

Anmelden / Registrieren
LesenForschungRadarAnlageframework

Quellen

  • tomshardware.com↗
  • the-decoder.com↗
Thema

美陆军AI网络代理

美国陆军网络司令部(ARCYBER)正开发名为“ECHOTRIBBLE”的自主AI网络攻击代理,旨在让软件在极少人类监督下自主导航敌方数字基础设施并“中和”目标。该项目处于原型征集阶段,引发安全专家对失控风险的强烈担忧:鉴于现有AI代理已在…

4 Schritte3 BeiträgeAktualisiert 2026-10-24 06:09

Wie es dazu kam

  1. 2026-06-24美国陆军网络司令部(ARCYBER)与网络融合创新中心(CFIC)举办线下协作活动,首次向业界展示ECHOTRIBBLE概念,演示AI代理如何像蠕虫一样在网络间自我复制并定位目标。cyberfic.org ↗
  2. 2026-10-19The Intercept获取内部文件,披露ECHOTRIBBLE项目要求AI代理具备“最小人类监督”下的自主行为能力,可完全禁用或删除威胁,且能穿透隔离网络。theintercept.com ↗
  3. 2026-10-20Tom's Hardware等媒体跟进报道,蒙特利尔大学David Krueger等专家警告,基于当前AI已知故障模式,授权自主系统执行“中和”任务极度危险,可能误伤美国水务设施等民用目标。tomshardware.com ↗

Was noch ungeklärt ist

  • 最终中标厂商是谁?其提供的AI模型是否经过针对军事场景的安全对齐测试?
  • 美国国防部是否会为ECHOTRIBBLE制定明确的交战规则(ROE),以界定“附带损害”的法律边界?
  • 在缺乏人类实时干预的情况下,如何技术上保证AI代理不会因幻觉或逻辑错误而攻击中立国或盟友的基础设施?

Auch in diesem Thread

  • US Army Solicits ECHOTRIBBLE Prototype: AI Agents to Autonomously Navigate and Neutralize Digital Threats2026-10-22
  • Leaked Documents Reveal US Army Cyber Command Developing 'ECHOTRIBBLE' Autonomous AI Attack Program2026-10-21
Vollständiges Thema lesen →
Recherche fortsetzen →