顯示具有 Sammy Azdoufal 標籤的文章。 顯示所有文章
顯示具有 Sammy Azdoufal 標籤的文章。 顯示所有文章

2026年2月20日 星期五

家中的特洛伊木馬:這次「木馬」甚至不是免費的

 家中的特洛伊木馬:這次「木馬」甚至不是免費的


特洛伊木馬的故事,幾個世紀以來都在提醒我們:有些「禮物」背後藏著敵人。如今這個隱喻顯得異常貼近現實:在許多家庭裡,那匹「木馬」不僅不再免費,還可能正在實時監視、聆聽,甚至繪製你私密空間的地圖。科技媒體 The Verge 近日報導的一起事件,就揭露了現代智慧家電如何變成特洛伊式的後門。

西班牙工程師 Sammy Azdoufal 原本並不想駭入全世界的掃地機器人。他只是想用 PS5 手把遠端操控新買的 DJI Romo,於是自己寫了遙控 App,並逆向分析 DJI 的通訊流程。沒想到,當他的程式連上 DJI 伺服器後,看到的不只是自己的機器,而是全球約 7,000 台 Romo 的即時資料,瞬間讓他變成無意間可以指揮成千上萬台掃地機器人的「王者」。

在測試中,他能遠端操控機器人、觀看即時鏡頭畫面、聽到麥克風收音,還能看著機器人建立 2D 平面圖,並用 IP 位址推估大致位置。他只抽出了自己的私密 token(讓伺服器認得「你可以看你的資料」的鑰匙),伺服器卻把別人的資料也一併交了出來。他形容「我的裝置只是資訊汪洋中的一台」,揭露了只要一扇門打開,就可能看到整片海洋。

在現場示範中,他的筆電每 3 秒就收到大量裝置的 MQTT 訊息:序號、正在清掃的房間、行進距離、是否回充電座、遇到的障礙等。短短 9 分鐘,他就盤點到 24 國、約 6,700 台裝置,累積超過 100,000 則訊息;若再加上同樣連到這些伺服器的 DJI Power 行動電源,可見裝置數甚至破萬。只要輸入 14 位數的序號,他就能查到媒體同事正在測試的那台 Romo,看到它在客廳清掃、電量 80%,並在不同國家遠端看著它把住宅格局正確建模。

Azdoufal 通報後,DJI 緊急修補,先是限制權限,讓他無法再操控別人的機器或看即時畫面與麥克風;隔天連他自己機器的資料也看不到了,顯示主要漏洞可能已補上。但這起事件仍引發對 DJI 資安與資料治理的質疑:一個工程師都能「撞」到成千上萬台裝置,如果是惡意攻擊者呢?再加上掃地機器人配備麥克風本身就讓人不安,他也直言「吸塵器上裝了麥克風真的很怪」。

DJI 後來承認,問題核心在於「後端權限驗證」,也就是裝置與伺服器之間以 MQTT 為基礎的通訊權限控管。公司表示 1 月底內部檢視時發現漏洞,2 月 8 日先推第一波修補,2 月 10 日再完成第二波更新,並稱已全數解決且不需使用者操作。DJI 也駁斥「不加密傳輸」的說法,強調 Romo 與伺服器之間以 TLS 加密傳輸。但研究者指出,就算「管道」加密,若缺乏細緻的 topic 權限控管,同一系統內的越權訂閱仍可能看到大量裝置訊息。TLS 擋得住竊聽,卻擋不住系統內部的權限失守。

Azdoufal 表示,目前仍有其他漏洞未完全修好,例如可在不輸入安全 PIN 的情況下看自己的 Romo 影像串流,另有更嚴重的問題他選擇暫不公開,DJI 則稱會在這週內處理。

這起事件的真正重點,不只是某一款產品的 bug,而是一種模式:許多「智慧」家電預裝了鏡頭、麥克風與雲端連線,被包裝成便利,實則可能成為監視工具。家中的特洛伊木馬,不再是被留在城門外的木馬,而是我們自己掏錢買回來、插上電、請進臥室與客廳的光鮮家電。這次,那匹木馬甚至不是免費的——它的代價,可能不是金錢,而是隱私。



The Trojan Horse in Our Homes: When the “Smart” Vacuum Costs More Than Money

 The Trojan Horse in Our Homes: When the “Smart” Vacuum Costs More Than Money


For centuries, the story of the Trojan Horse has served as a warning about gifts that carry hidden enemies. Today, that metaphor feels disturbingly literal: in many homes, the “horse” is no longer free, and it may be watching, listening, and mapping our private spaces in real time. A recent report by The Verge about a security researcher’s accidental discovery inside DJI’s Romo robot vacuum illustrates how modern smart devices can become Trojan‑style backdoors into our lives.

Spanish engineer Sammy Azdoufal did not set out to hack the world’s robot vacuums. He simply wanted to control his newly bought DJI Romo with a PS5 controller, so he wrote his own remote‑control app and reverse‑engineered DJI’s communication flow. When his app connected to DJI’s servers, however, it did not see just his device. Instead, it received live data from roughly 7,000 Romo units around the world, suddenly turning him into an unintended “commander” of thousands of strangers’ household robots.

In his tests, Azdoufal was able to remotely move the vacuums, view live camera feeds, and even hear audio from their microphones. He could watch each robot build detailed 2D floor plans of homes and use IP addresses to approximate their locations. He described extracting only his own private authentication token—the key that tells the server “you are allowed to see your data”—yet the server handed over other people’s data as well. “My device was just one in an ocean of information,” he said, revealing how easily one user’s access could bleed into everyone else’s.

During a live demonstration, his laptop received MQTT messages from thousands of devices every three seconds: serial numbers, which room was being cleaned, distance travelled, whether the robot was returning to its dock, and what obstacles it had encountered. In just nine minutes, he catalogued about 6,700 units across 24 countries, logging more than 100,000 messages. When he included DJI Power power banks connected to the same servers, the visible device count exceeded 10,000. By typing in a 14‑digit serial number, he could pull up a colleague’s Romo in another country, see it cleaning the living room, check its 80% battery, and watch it map the home layout in real time.

After Azdoufal alerted the media, DJI moved quickly. By Tuesday, he could no longer control other people’s Romos or view live video or microphone feeds. By Wednesday morning, even his own device disappeared from his scanner, suggesting that DJI had closed the main leak. Yet the episode raised serious questions about DJI’s security and data governance: if a curious engineer could stumble on a flaw exposing thousands of devices, what could a malicious actor do? And why does a vacuum cleaner need a microphone at all?

DJI later acknowledged that the core issue lay in backend permission validation—how devices and servers manage access via MQTT‑based communication. The company said it had internally detected the vulnerability in late January, rolled out an initial patch on February 8, and completed a second update on February 10, claiming the problem was fully resolved without user action. DJI also denied that data was transmitted unencrypted, insisting that Romo communicates with servers over TLS. However, researchers point out that even with encrypted channels, poor topic‑level permission controls can still allow an authorized client to see messages from many unrelated devices. Encryption protects the pipe, not the permissions inside the system.

Azdoufal noted that other vulnerabilities remain, such as being able to view his own Romo’s video stream without entering a security PIN, and at least one more serious flaw he chose not to disclose. DJI said it would address these issues within the week.

The real story here is not just a bug in one product line; it is a pattern. Many of today’s “smart” home devices come pre‑installed with cameras, microphones, and cloud connectivity, sold as conveniences but capable of functioning as surveillance tools. The Trojan Horse in our homes is no longer a wooden gift left at the gate; it is a sleek, branded appliance we willingly plug in ourselves, pay for, and invite into our bedrooms and living rooms. This time, the horse is not even free—and its price may be measured not in gold, but in privacy.