독서리서치레이더투자 프레임워크
로그인 / 회원가입
← 읽기로 돌아가기
뉴스주제 · 美陆军AI网络代理구조적2 분 소요DisconfirmAI 편집부

Leaked Documents Reveal US Army Cyber Command Developing 'ECHOTRIBBLE' Autonomous AI Attack Program

US Army Cyber Command is developing ECHOTRIBBLE, a program to deploy autonomous AI agents that can navigate networks and 'neutralize' adversary targets with 'minimal human oversight.'

2026-10-19 18:00 GMT+8

Documents obtained by The Intercept reveal that the U.S. Army Cyber Command (ARCYBER) is developing a nascent program codenamed "ECHOTRIBBLE," designed to use autonomous artificial intelligence agents to hack its adversaries. The files indicate the military aims to harness the current trend of AI agents—software leveraging large language models to autonomously accomplish tasks—and convert them into tools of war capable of navigating infrastructures and identifying and "neutralizing" threats.

According to an ARCYBER document circulated to potential contractors, ECHOTRIBBLE would function almost like a computer worm, silently navigating foreign networks and hopping from device to device until it finds the Army's quarry. It is designed to hit targets in the cloud as well as those on "air-gapped" systems disconnected from networks entirely. Once its task is complete, the agents would effectively self-destruct, ideally without having been noticed by the target.

While the software can be programmed with initial "predefined triggers" dictating what and where to attack, the documents state it "must operate with minimal human oversight" and be capable of "Self-Directed Behavior." A presentation to vendors provided a hypothetical example: once uploaded to a laptop, the software identifies other potential targets on the same network and produces replicas tailored for iPads, Apple desktops, and Windows 11 PCs.

Experts warn that this move comes amid constant revelations of rogue behavior by AI agents, including unauthorized breaches of government and corporate systems. David Krueger, a machine learning professor at the University of Montreal, stated that given known failure modes, making autonomous AI systems with a mandate to "neutralize threats" is incredibly dangerous. Hyrum Anderson, CEO of AI security firm Abundant, raised concerns that such agents might route attacks through civilian critical infrastructure, causing collateral damage to public utilities or small businesses.

로그인 / 회원가입
독서리서치레이더투자 프레임워크

출처

  • theintercept.com↗
주제

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 단계3 건업데이트됨 2026-10-24 06:09

현재까지의 과정

  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 ↗

아직 해결되지 않은 문제

  • 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?

이 스레드의 다른 내용

  • US Army Solicits ECHOTRIBBLE Prototype: AI Agents to Autonomously Navigate and Neutralize Digital Threats2026-10-22
  • US Army reportedly developing autonomous AI cyberattack agents, experts warn of loss of control2026-10-20
전체 주제 읽기 →
리서치 계속하기 →