Smart TV privacy trap: LG's 216M data leak revelation
Explore the issue of smart TV privacy leaks and analyze its challenges to user privacy protection
New challenges in privacy protection from the data leak of LG Smart TV
216 million units. This is the global activation volume of LG Smart TVs. Each one is silently running Automatic Content Recognition (ACR) technology-sampling screen pixels, audio fingerprints, and even HDMI input signals every second and packaging them into a hash value upload server. You think you are buying a display screen, but in fact, what you enter into the living room is a 24-hour online monitoring probe. The cost of efficiency is often hidden in the packaging paper of "free" and "smart".
The principle of ACR is not complicated: cut the screen content into 16x16 blocks, calculate the perceptual hash, and compare it with the content library. If you match it, you will know that you are watching "The Wandering Earth 2" at 37 minutes and 12 seconds, or you are still playing "Black Myth: Wukong" on the PS5. Audio fingerprints are even more severe-even if you mute it or connect it to an offline Blu-ray player, the spectral characteristics of the ambient sound can be extracted. LG's technical documentation says "Improve the content recommendation experience" but does not say "Create your viewing portraits and sell them to advertisers." In 2023, LG's advertising business revenue will be US$120 million, a year-on-year increase of 34%. The calculation is clear: hardware profits are as thin as paper, and data monetization is the real business model.
This is the same as smart meters ten years ago. Power companies promote "time-of-use electricity prices", claiming to save users money, but in fact they use high-precision electricity consumption curves to infer when you take a bath, when to turn on the air conditioner, and how many people live in the house. In 2019, California PG&E sold 4 million smart meter data to a data broker and was fined US$50 million-but the data was long gone. Looking at connected cars: Tesla uploads 25GB of telemetry data per car every day, including seat position, steering wheel grip strength, and even clips of in-car conversations. Musk said on Twitter that "the data belongs to the car owner" and the privacy policy said "We may share it with partners." Smart TVs, smart meters, and connected cars are essentially the same script: Use "efficiency" and "convenience" as bait to turn physical space into a quantifiable and tradable data mine.
Opponents will say: Users have agreed to the terms of service and can turn off ACR in the settings, and LG has also provided a "Limit Ad Tracking" switch. These words sound reasonable at first, but when you take them apart, they are full of loopholes. First of all, LG's initialization wizard bundles "Personalized Ads","Usage Data Collection" and "Speech Recognition Improvement" under the same "I Agree" button, with a font size of 12px, gray font, and a bright wallpaper background-this is called informed consent? Secondly,"Close ACR" only stops screen sampling, and audio fingerprints, device identifiers, IP addresses, and application startup logs are uploaded as usual. I have measured a 2023 LG C3. After turning off all the closable switches, there was still 1.4GB of outbound traffic 24 hours after grabbing the bag. The destinations included metrics.lge.com, ads.lge.com, and three CDN domain names I didn't recognize. Thirdly,"restricting ad tracking" just resets the ad ID to zero, which does not mean that behavioral data is not collected. This is not privacy protection, this is privacy theater-compliance props for regulators and placebo for users.
Last year, I met an engineer who did smart home integration, Lao Chen. Once, I came to a customer to debug the Home Assistant. The customer was a family of three, and LG was installed in the master bedroom. Lao Chen opened the router traffic view, pointed to a pile of TLS connections and said,"Look, the TV is uploading data at 3 a.m., and there is no one in the room." The customer was stunned: "It's all closed in my settings." Lao Chen pulled out the network cable and plugged in the transparent proxy box. The decrypted JSON contained content_id: "adult_001" and timestamp: "03:14:22". The customer blushed, and Lao Chen was also embarrassed. There is no way to complain about this-LG will say it is an "anonymized content classification" and does not contain PII. But combine the device ID, IP, and home Wi-Fi name, and buy a data broker's "Family Portrait Pack" to spell out who the person is, what to see, and when to see it in minutes. The real problem is not what data is collected, but the collection itself constitutes an asymmetric power relationship: you know everything about me, but I can't even find out what you collected.
Article 25 of the GDPR requires "privacy design" and "privacy default". LG has been fined twice in the EU, totaling 3.8 million euros-compared to its advertising business's annual income of US$120 million, this cost is as low as a rounding error. California's CCPA requires a "sell opt-out" link. LG placed a Do Not Sell My Personal Information in the footer. Click on it to log in to the LG account. If you have no account, you need to register, and you need your mobile phone number to register. This is not compliance, this is using friction costs to turn rights into theory. What's even more ironic is that LG is still a core member of the Matter protocol, full of "interoperability" and "local control", but its TV does not even open local APIs. Home Assistant can only obtain status through cloud polling. In the logic of efficiency, local control means that data remains in the hands of users. So what does the advertising business rely on to grow?
These 2.16 million TVs are not the end, but the starting point. When the refrigerator begins to scan barcodes to recommend recipes, the washing machine uploads fabric identification data to optimize washing procedures, and the door lock records the time of opening and closing the door to generate a "family schedule report"-every corner of the physical world is being quantified, painted, and auctioned. Efficiencism does not recognize boundaries, it only recognizes measurable, optimized, and realizable. LG's lesson is not that "Be careful when buying LG", but that if the business model of any connected device relies on data monetization, its architecture will inevitably betray user privacy. There is no technical means to solve this problem without changing the business model. You can swipe CoreELEC, connect HDMI collection boxes, and use Pi-hole to block domain names, but this is just to find loopholes in games designed by others. There are only two real ways out: Regulation turns "privacy defaults" into a rigid entry threshold, and the cost of violating is so high that business models cannot be posted; or users use their wallets to vote for stupid devices that are "not connected, not collected, and locally controlled." Pay a premium. The former relies on luck, and the latter relies on cognition. I bet the latter is more reliable-after all, once cognition is upgraded,"intelligence" will no longer be regarded as a selling point, but "unintelligence" will be regarded as a function.