選單
中文 (繁體)

This sucks: intimate photos taken by robot vacuums were leaked online

Even if you’re not a tech geek, you likely have a few “smart” devices at home. A vacuum cleaner, every modern homemaker’s best friend, is probably one of them. A robot vac can sweep floors, hover up dust, pet hair — but also your data.

Privacy policies of most humming assistants state that the manufacturer can collect information from you to improve their products and services, provided you opt in. But who doesn’t want their vacuum to become better in dodging dog poop, avoiding wires, and performing other circus tricks such as not getting stuck under the bed? And who reads the privacy policy before clicking ‘accept’ anyway? Besides, one can be under the impression that an impartial algorithm will process the collected data or, at the very least, that this data will stay with the manufacturer and not be spilled all over the internet — so no need to worry.

iSpy: how a robot vacuum leaked intimate pics

But this assumption is only partially true. While AI is definitely on the course to replace humans, it is not there yet. MIT Technology Review revealed last month that the leading robot vacuum cleaner manufacturer, iRobot, hired third party contractors to sift through and label the raw data collected by the development version of its Roomba j7 series robot. iRobot shared the data with a startup called Scale AI, which, in turn, shared it with independent subcontractors from Venezuela, who in turn shared it on… Facebook and Discord.

These low-paid workers were tasked with labeling furniture and other objects that Roomba’s cameras have captured. The majority of the 15 images obtained by MIT Technology Review from private social media groups were quite innocuous: they featured doorways, shelves, kitchen cabinets, lights, countertops etc. But several photos were way more damning. A series of images taken from different angles showed a woman sitting on a toilet seat — in some of the stills her face is clearly visible. Another image featured a child sprawled on the floor and staring directly at the camera, his face not blurred as well.

The leak set off a blame game. iRobot accused Scale AI of breaching the contract by allowing the images to be posted. Scale AI, for its part, accused Venezuelan workers of violating their contracts.

One could argue that iRobot fell victim to a bad subcontractor. Indeed, if the labellers hadn’t made the blunder of posting the photos on social media, nothing would have come to light. But the problem goes much deeper than a bunch of crowdsourced workers that are yet to grasp the concept of privacy — it’s about how much data tech companies glean from us and who they share it with.

The issue of consent

In its defense, iRobot claimed that the model that captured the compromising images was not sold in stores, but was only distributed among “paid collectors and employees.” Those people, iRobot said, had all consented to their data, including videos, being used for AI training. iRobot, though, did not say whether the volunteers knew that their intimate data would be seen by fellow humans. The consent form only mentioned that the data would be shared with “service providers,” according to the MIT Review.

iRobot admitted that it had shared visual data collected by its vaccums — over 2 million images in total — with multiple contractors and not only Scale AI. According to iRobot, before handing the data over, it filters out images where a user is naked or otherwise compromised. The company did not say why the image of a woman on the toilet was not removed, however.

What does your vacuum know about you

iRobot maintains it does not spy on regular consumers. But, technically, nothing stops it from collecting large amounts of precise information from unsuspecting users’ homes.

First of all, robot vacuums collect this data when they map out your home — that involves drawing a detailed map of all the rooms in your house during the initial cleaning. The map helps the robot understand the layout of the rooms so that it does not clean the same spot twice and bump into corners. The more advanced a mapping technology is, the more accurate the map, and the more efficient the cleaning. In the case of iRobot’s j7 model, the vacuum can collect this data thanks to a front-facing AI-powered camera, a step up from gyroscope and accelerometer-based mapping technology used by cheaper models. Having been fitted with a camera not only gives a robot vacuum an edge over its blind competitors, but also carries obvious privacy risks — since it is now able to see what’s going round and take photos.

Thus, iRobot’s Roomba j7+ model can recognize and take pictures of pet waste and cables by itself and share them with a smartphone app, if you opt in. You can also opt in to share your data with iRobot; in this case, the images will be encrypted and sent to the company, where human employees will examine them. As far as images of people go, the vacuum’s software is supposed to automatically shut off the camera if a human or a photo of a human is in its view. But, as we’ve seen from the MIT Review’s report, safeguards do not always work.

Robot vacuums have learnt to identify and avoid pet waste
Photo: YoonJae Baik/Unsplash

Moreover, as the saying goes, appetite (for data) comes with eating, or rather, vaccuming and mopping in this case.

The existing technology allows vacuuming robots to get plenty of valuable insights into your life that can come handy to third parties, such as advertisers. And it’s not just about the size of your house, whether you have two bedrooms or one, or whether you have a pet that poops all over the place — although, this is already, perhaps, more than you want strangers to find out. The vacuum cleaner can also get information about your daily routine. With iRobot OS, a cloud-based operating system for robot vacuums, you can set up the iRobot vacuum cleaner so that it automatically starts cleaning when you leave the house and stops when you return. The robot can also be paired with voice assistants like Amazon Alexa, Apple Siri, and Google Assistant so that it responds to your voice commands.

Since iRobot OS is cloud-based, all this information can potentailly be accessed by rogue employees or shared with third parties, such as law enforcement. And the amount of information vacuum robots can amass is nothing short of impressive. iRobot CEO Colin Angle told the Verge in May that the j7 vacuums have so far detected over 43 million objects in people’s homes and learnt to identify and avoid 80 individual objects. This basically means that the manufacturer sits on a treasure trove of data. The question is, will he open it and for whom?

Amazon and iRobot — a match made in privacy hell?

Perhaps, it’s not a coincidence that once iRobot improved its navigation system and revamped its cloud-based software, it sparked Amazon’s interest. In August this year, the companies signed a merger deal, according to which Amazon will pay $1.7 billion for iRobot. The deal, which is yet to be approved by regulators, set off a privacy firestorm.

A number of data privacy organizations, including the Electronic Frontier Foundation, urged the US regulator to derail the acquisition, arguing that the potential comingling of Amazon and iRobot’s data would “jeopardize consumer privacy.”

“Linking iRobot devices to the already intrusive Amazon home system incentivizes more data collection from more connected home devices, potentially including private details about our habits and our health that would endanger human rights and safety,” the letter read.

In fact, iRobot has been entertaining the idea of sharing data with Big Tech long before settling for Amazon. In 2017, iRobot CEO Angle said that his company could share maps of customers’ homes for free with the likes of Amazon, Apple, and Google with customer consent in the future.

It should be noted that Amazon’s pending acquisition of iRobot is just part of a bigger, a rather worrisome trend.

Last year, Google bought fitness tracking company Fitbit for $2.1 billion, sparking similar privacy concerns. In the end, Google agreed to store Fitibit data separately and not use it for Google ads for at least 10 years so that the acqusition deal passes muster with the European Commission. Whether you should trust Google on this — is up to you, but as history shows its promises as far as privacy is concerned are not set in stone

Sensors get better — privacy issues get worse

On par with camera-based vacuuming robots, there are also laser-based robots that use a spinning invisible laser to glean detailed information about the size of the rooms and obstacles. This cutting-edge technology is called LiDAR, which stands for Light Detection and Ranging, and is considered to be the “gold standard for vacuum cleaning navigation”. Laser-based robots tend to be more accurate than their camera-based counterparts, and are typically much faster. And while they can’t capture as much visual data, they also pose a privacy threat, albeit of a different nature.

Researchers have found that devices that use LIDAR can be hacked to collect audio data from homes. Even though they do not have physical microphones installed, these robot vacuums can be repurposed “to spy on conversations and potentially reveal private information,” the researchers from the University of Maryland and Singapore concluded. In the paper called Spying with Your Robot Vacuum Cleaner: Eavesdropping via Lidar Sensors, the researchers described how they used a Xiaomi Roborock vacuum cleaner to successfully record speech and music with over 90% accuracy.

Robot vacuums have advanced sensors that help them scan the surroundings and avoid obstacles, such as cords
Photo: Elena Mozhvilo/ Unsplash

It’s safe to assume that sensors will get better with time as the demand for greater precision and efficiency while cleaning will only increase. Who wants their vacuum to trip on cords, ram a dog bowl, or scatter pet waste? That means that vacuums’ ability to collect data from homes will also increase. Besides, for the sake of more accuracy, manufacturers may consider installing multiple cameras at once (e.g. ceiling and front), or using both a laser and a camera for navigation — that will potentially enable them to collect even more data.

Balancing out the risks

Internet-connected devices pose a serious threat to privacy, and robotic vacuum cleaners are no exception. The fact that, unlike voice assistants and smart doorbells, which are stationary devices, vacuum cleaners can move and in some cases snap pictures of the inside of your home. Which makes them even more dangerous privacy-wise.

While manufacturers may assure you that they take every precaution to not let confidential user information leak, safeguards can fail, as we’ve seen in the case of Roomba. That’s why the choice is often between privacy, security, on the one hand, and convenience on the other.

The most privacy-friendly option will be to “dumb down” your smart home and replace a robot vacuum cleaner with a good old-fashioned corded vacuum which you would need to manually plug into a socket and drag across the floor. While privacy-friendly and cost-efficient, this option is hardly labor- or time-saving. And if you’re willing to sacrifice some of your privacy and security for convenience, you should be aware of the risks, and follow some basic rules:

  • Opt for trustworthy vendors with after-sale support so your data is less at risk from a potential oversight

  • Read the privacy policy and Terms of Services (ToS) and carefully tick privacy settings — while some features may not work unless you agree to a certain degree of data sharing, do not agree to more data sharing than is necessary for your device to function

  • We usually recommend using DNS filtering to block access to ad servers and tracking sites. However, when data collection is part of the device’s core functionality, it cannot be restricted without undermining that device’s operation — this is true not only for vacuum cleaners, but also for other IoT devices, such as smart speakers.

喜歡這篇文章嗎?
19,180 19180 使用者評論
極好的!

AdGuard for Windows

Windows 版 AdGuard 不只是廣告封鎖程式,它是集成所有讓您享受最佳網路體驗的主要功能的多用途工具。其可封鎖廣告和危險網站,加速網頁載入速度,並且保護兒童的線上安全。
透過下載該程式,您接受授權協定的條款
閱讀更多
19,180 19180 使用者評論
極好的!

AdGuard for Mac

Mac 版 AdGuard 是一款獨一無二的專為 MacOS 設計的廣告封鎖程式。除了保護使用者免受瀏覽器和應用程式裡惱人廣告的侵擾外,應用程式還能保護使用者免受追蹤、網路釣魚和詐騙。
透過下載該程式,您接受授權協定的條款
閱讀更多
19,180 19180 使用者評論
極好的!

AdGuard for Android

Android 版的 AdGuard 是一個用於安卓裝置的完美解決方案。與其他大多數廣告封鎖器不同,AdGuard 不需要 Root 權限,提供廣泛的應用程式管理選項。
透過下載該程式,您接受授權協定的條款
閱讀更多
19,180 19180 使用者評論
極好的!

AdGuard for iOS

用於 iPhone 和 iPad 的最佳 iOS 廣告封鎖程式。AdGuard 可以清除 Safari 中的各種廣告,保護個人隱私,並加快頁面載入速度。iOS 版 AdGuard 廣告封鎖技術確保最高質量的過濾,並讓使用者同時使用多個過濾器。
透過下載該程式,您接受授權協定的條款
閱讀更多
19,180 19180 使用者評論
極好的!

AdGuard 內容阻擋器

AdGuard 內容阻擋器將消除在支援內容阻擋器技術之行動瀏覽器中的各種各類廣告 — 即 Samsung 網際網路和 Yandex.Browser。雖然比 AdGuard for Android 更受限制,但它是免費的,易於安裝並仍提供高廣告封鎖品質。
透過下載該程式,您接受授權協定的條款
閱讀更多
19,180 19180 使用者評論
極好的!

AdGuard 瀏覽器擴充功能

AdGuard 是有效地封鎖於全部網頁上的所有類型廣告之最快的和最輕量的廣告封鎖擴充功能!為您使用的瀏覽器選擇 AdGuard,然後取得無廣告的、快速的和安全的瀏覽。
19,180 19180 使用者評論
極好的!

AdGuard 助理

AdGuard 桌面應用程式的配套瀏覽器擴充功能。它為瀏覽器提供了自訂的元件阻止的功能,將網站列入允許清單或傳送報告等功能。
19,180 19180 使用者評論
極好的!

AdGuard DNS

AdGuard DNS 是一種不需要安裝任何的應用程式而封鎖網際網路廣告之極簡單的方式。它易於使用,完全地免費,被輕易地於任何的裝置上設置,並向您提供封鎖廣告、計數器、惡意網站和成人內容之最少必要的功能。
19,180 19180 使用者評論
極好的!

AdGuard Home

AdGuard Home 是一款用於封鎖廣告 & 追蹤之全網路範圍的軟體。在您設置它之後,它將涵蓋所有您的家用裝置,且為那您不需要任何的用戶端軟體。由於物聯網和連網裝置的興起,能夠控制您的整個網路變得越來越重要。
19,180 19180 使用者評論
極好的!

AdGuard Pro iOS 版

除了在 Safari 中之優秀的 iOS 廣告封鎖對普通版的用戶為已知的外,AdGuard Pro 提供很多功能。透過提供對自訂的 DNS 設定之存取,該應用程式允許您封鎖廣告、保護您的孩子免於線上成人內容並保護您個人的資料免於盜竊。
透過下載該程式,您接受授權協定的條款
閱讀更多
19,180 19180 使用者評論
極好的!

AdGuard for Safari

自 Apple 開始強迫每位人使用該新的軟體開發套件(SDK)以來,用於 Safari 的廣告封鎖延伸功能處境艱難。AdGuard 延伸功能可以將高優質的廣告封鎖帶回 Safari。
19,180 19180 使用者評論
極好的!

AdGuard Temp Mail

免費的臨時電子郵件地址產生器,保持匿名性並保護個人隱私。您的主收件匣中沒有垃圾郵件!
19,180 19180 使用者評論
極好的!

AdGuard Android TV 版

Android TV 版 AdGuard 是唯一一款能封鎖廣告、保護隱私並充當智慧電視防火墻的應用程式。取得網路威脅警告,使用安全 DNS,並受益於加密流量。有了安全性和零廣告的使用體驗,使用者就可以盡情享受最喜愛的節目了!
已開始下載 AdGuard 點擊箭頭所指示的檔案開始安裝 AdGuard。 選擇"開啟"並點擊"確定",然後等待該檔案被下載。在被打開的視窗中,拖曳 AdGuard 圖像到"應用程式"檔案夾中。感謝您選擇 AdGuard! 選擇"開啟"並點擊"確定",然後等待該檔案被下載。在被打開的視窗中,點擊"安裝"。感謝您選擇 AdGuard!
在行動裝置上安裝 AdGuard