A Chinese-speaking operator is said to have pointed autonomous AI agents at company websites and let them attempt break-ins with almost no hands-on help — and a security researcher was able to watch the campaign unfold because the operators left their control panels open on the internet.
The finding, reported by The Hacker News in October 2026, puts the scale at more than 1,200 sites probed and websites belonging to more than 100 companies reached, including farm-equipment maker Deere and one unnamed appliance manufacturer. All of those figures, along with the characterisation of the operator's language and working style, are attributed to the reporting researcher rather than independently confirmed.
The distinction that matters most is also the one most easily blurred: reaching a site is not the same as compromising it. The public write-up does not claim 100 breaches, and no count of sites actually penetrated has been disclosed.
The novelty: the attacker was software
What sets this campaign apart is not the target list but the delegation. Intrusion decisions appear to have been handed to automated agents rather than made by a person at a keyboard, with human involvement described as minimal — though how much the operator stepped in on any given attempt was not disclosed.
For defenders, that shift outweighs the raw counts. Machine-driven activity does not keep human hours, pause between attempts the way a person would, or repeat itself along the patterns detection rules are typically tuned around. Indicators of compromise may also land faster than a small analyst team can triage them, because the attack loop itself runs at machine speed.
The hook: dashboards on the open internet
This campaign surfaced not through a breach notification but through a basic OPSEC lapse. The control panels used to manage the operation were exposed on the open internet, and the researcher behind the finding used that misconfiguration as a window into the campaign from the outside.
The symmetry is the story's real hook. An automated intrusion capability — machine-speed, aimed at more than a thousand targets — was undone by the same class of forgotten admin interface that attackers themselves exploit. Exposure management cuts both ways: the panel that lets an attacker in can also be what gives them away.
What remains unclear
Several questions remain open, and they are the ones worth tracking: what "reached" specifically means in this case — exploitation, a foothold, or mere enumeration; how many of the 1,200-plus targeted sites were actually compromised; what data or access was taken from the more than 100 companies whose sites were reached; how the agents technically carried out the work; and how much the operator intervened per attempt. Follow-up reporting on any of these points would be the story's real substance.
Why it matters
If this campaign is any indication, the operational lesson sits where the campaign itself collapsed: on the exposed surface. Teams running public-facing sites should be inventorying and closing internet-exposed admin interfaces behind VPN or identity-aware access, scanning for forgotten panels on a schedule rather than by spot check, logging admin-panel access so an unexpected login is visible early, and preparing detection and triage workflows to absorb bursts of automated probing — because the advice for a machine-speed attacker is the same advice the attackers failed to follow themselves.
Source: The Hacker News, October 2026.
一名安全研究員稱，一名講中文的犯罪者把自主 AI 代理指向企業網站，幾乎不親自動手就讓它們嘗試入侵——而研究員之所以能目睹整個行動展開，是因為操作者把自己的控制面板留在了互聯網上。
該發現由 The Hacker News 於 2026 年 10 月報道，規模為探測逾 1,200 個網站、觸及逾 100 家公司的網站，其中包括農機製造商 Deere，以及一家未具名的家電製造商。上述所有數字，連同對操作者語言及工作方式的描述，均出自報道中的研究員，並非獨立證實。
最重要、也最容易被混淆的界線是：觸及一個網站並不等同於成功入侵。公開文章並未聲稱有 100 宗入侵，實際被攻陷的網站數目亦未有披露。
新奇之處：攻擊者是軟件
本次行動與別不同之處，不在目標名單，而在於決策的下放。入侵決策似乎交由自動化代理、而非鍵盤前的人來作出，人的參與被形容為極少——惟操作者在每次嘗試中介入多少，未有披露。
對防禦者而言，這個轉變比原始數字更值得注意。由機器驅動的活動不會跟隨人的作息，不會像人一樣在嘗試之間停頓，也不會沿著偵測規則通常據以調校的重複模式行事。入侵指標（IoC）也可能比小型分析團隊消化得更快，因為攻擊循環本身以機器速度運行。
關鍵：控制面板暴露在公開互聯網
本次行動之所以浮現，不是因為一則戲劇性的入侵通知，而是因為一次基本的 OPSEC 失誤。用於管理整個行動的控制面板暴露在公開互聯網上，發現事件的研究員藉著這項配置錯誤，從外部打開了觀察行動的一扇窗口。
對稱性正是這則報導的核心。一套自動化入侵能力——以機器速度運行、瞄準逾千個目標——最終被攻擊者自己常用的那類被遺忘管理介面擊倒。暴露面管理是雙向的：讓攻擊者得以進入的面板，同樣可以是令他們現形的東西。
仍然未清楚的地方
以下問題仍然懸而未決，亦是值得追蹤的方向：「觸及」在此究竟指什麼——成功利用、取得立足點，還是僅僅完成掃描列出；1,200 多個目標網站中有多少實際被入侵；那 100 多家被觸及網站的公司中，有哪些資料或權限被取走；代理以什麼技術方式完成入侵；以及操作者每次介入的程度。以上任何一點的跟進報道，才是這則新聞真正的實質內容。
為什麼重要
如果本次行動可作參考，其營運教訓恰恰落在行動自身崩塌之處：暴露的攻擊面。營運公開網站的團隊，應把互聯網上暴露的管理介面盤點出來並關閉，改以 VPN 或身分感知存取保護；應定期自動掃描遭遺忘的面板，而非靠一次性人手覆核；應記錄並監控管理面板的登入，令異常嘗試盡早可見；並應讓偵測與分類流程能夠承受自動化探測的爆發流量——因為針對機器速度攻擊者的建議，正是攻擊者自己沒有做到的那套建議。
資料來源：The Hacker News，2026 年 10 月。