Samsung TV exposed: hidden neural net tracks viewing, audio classification, and facial recognition code found in firmware

Samsung's Smart TVs Run Hidden Neural Networks for Viewing Surveillance

Samsung's recent firmware includes region-specific convolutional neural networks that track user behavior within apps—monitoring which videos are watched, which ads are skipped, and video playback progress. The discovery, surfaced by security researcher Log4j, also found code for facial recognition that identifies people appearing on screen, though the mechanism for this detection appears to operate on video content rather than through a built-in camera (most modern Samsung TVs lack cameras; only select models from 2013 to 2015 included integrated cameras).

The TV also classifies audio it captures before reporting findings back to Samsung. Samsung has publicly marketed these capabilities as "Vision AI" features—specifically pet care and family care alerts that use sound sensors and on-device AI to detect barking or crying, triggering smart home notifications. Event counter documentation in the firmware aligns with these advertised features.

The broader tension is between disclosed functionality and the user experience cost. Samsung allows these features to run continuously in the background, slowing the television's overall performance. Users cannot fully disable the surveillance features even if they accept the premise of smart home monitoring.

The TV industry operates under substantially weaker privacy pressure than adjacent markets. Apple has invested heavily in privacy-forward messaging around its devices, and Meta disclosed that its camera glasses include indicator lights. Smart TV manufacturers have largely avoided similar scrutiny, operating without the same regulatory or reputational constraints.

The absence of a privacy-first TV alternative remains notable. Despite charging premium prices for other categories—gold Apple Watches at $10,000, foldable phones with multiple screens—Apple has not entered the television market. Such a product, if built with the company's design sensibility and privacy commitments, would likely differentiate sharply against ad-supported, feature-laden competitors. The core barrier appears to be business model alignment: a high-margin, privacy-respecting TV does not fit Apple's revenue model as cleanly as it might for a company willing to charge a premium specifically for the absence of tracking and advertising.