Cybersecurity bounty hunter says Samsung TVs use AI to record audio and screenshots for ads

A cybersecurity expert, known as log4j, discovered that Samsung TVs use artificial intelligence to record audio and capture screenshots for targeted advertisements. This was revealed after log4j began noticing unusual ads on their 2025 Samsung TV and started reverse-engineering the Tizen operating system. The investigation led to the disclosure of extensive user telemetry, including packet logging, audio categorization, screenshots, and cross-device identification, all aimed at improving targeted ads.

The reverse engineer expressed concern over the specific nature of the ads on their Samsung TV and chose to document their findings. The process was challenging due to AES-encrypted firmware and the absence of keys, requiring dismantling the device and using UART/JTAG methods. The findings were alarming, with log4j referring to them as "PRIVACY HORROR."

They theorized that the sluggish performance of Samsung TVs was due to extensive post-processing of user data, including audio, screenshots, and cross-device tracking IDs, along with computer vision processing. Log4j also discovered a region-specific convolutional neural network designed for user behavior tracking, observing elements such as watched videos, skipped ads, and video progress.

The hacker even found that the TV could locate and open a connected device's microphone for five seconds before ceasing. OpenAI tools were used during the process, but the company banned log4j for cyber abuse, to which the reverse engineer claimed they only utilized the tools for local reverse engineering, not for exploitation or violating terms of use.

The incident was not unique to Samsung, as other brands like LG also faced accusations of spying on users. Log4j is currently working on a comprehensive write-up detailing their findings and how others can replicate the process.