Security Researcher Claims Samsung Smart TVs Collect Audio and Screenshots for Ads
A cybersecurity researcher, operating under the handle @log4j, has claimed to have gained root access to his 2025 Samsung smart TV and discovered what he alleges are privacy-invasive features. The researcher asserts that the TV’s onboard AI collects audio and screenshots for targeted advertising, even capable of distinguishing between dog barks and infant cries. Following his public disclosures, @log4j reported being banned from OpenAI for “cyber abuse.” A full technical report has not yet been released, and neither Samsung nor OpenAI have publicly commented on the specific case as of this report.
Deep Dive into TV’s Alleged Data Collection
The researcher’s investigation began after he found the advertising on his TV to be “suspiciously accurate.” After obtaining a root shell on his 2025 Samsung TV, he detailed in a series of posts that the firmware, encrypted with AES, proved difficult to analyze. He stated he had to utilize a zero-day exploit to gain access and is still compiling a comprehensive write-up.
These claims are currently unilateral. Tom’s Hardware compiled the researcher’s posts on October 25th. Concurrently, Hugging Face engineer ngxson reportedly tested a microphone on a Flipper Zero, observing it activate for approximately 5 seconds before stopping. However, this device had been rooted and had multiple settings disabled, suggesting its behavior might differ from an unmodified unit.
On-Device AI and Cross-Device Tracking
According to the researcher’s posts around October 24th, the TV allegedly uses its own hardware for post-processing, with telemetry encompassing audio, screen captures, and cross-device advertising identifier (AID) tracking. He claims that computer vision and AI models run locally on the TV, rather than sending all data to the cloud for analysis. This effectively transforms the living room screen into an “on-device analysis” node. On the same day, he specifically pointed to a regional convolutional neural network (CNN) that he believes tracks in-app behavior, such as YouTube viewing habits, skipped ads, and playback progress.
Microphone Access Beyond the TV
Even for models without a built-in microphone, the researcher alleges the system attempts to access microphones from connected soundbars or Bluetooth devices, routing the audio to Bixby. This suggests concerns extend beyond the TV’s own microphone, potentially encompassing the unauthorized use of audio input from external devices in the vicinity.
Questioning Privacy Settings
Many users might assume that disabling Automatic Content Recognition (ACR) would halt data collection. However, on the evening of October 25th, the researcher cautioned that disabling ACR in the menu does not necessarily stop all telemetry. He indicated that some checks are either implemented incorrectly or are absent altogether, reiterating that his full report is still in preparation and urging journalists not to over-interpret his reverse-engineering snippets.
This caveat is significant. The presence of counters or code paths does not automatically equate to user notifications or the transmission of entire recordings. The researcher himself admits he has not yet fully investigated the lifecycle details. What is currently confirmed is his assertion that menu toggles may not be entirely trustworthy, but it remains unconfirmed what data, if any, is actually transmitted by unmodified Samsung TVs.
Audio Classification and Transparency Concerns
The researcher also claimed that the TV classifies heard audio and reports it to Samsung, citing preference settings, dog barks, and infant cries as examples within the telemetry context. The core of the controversy appears to be less about the ability to detect dog barks and more about the transparency of reporting and advertising tracking.
In contrast, Samsung’s official narrative, through features like Vision AI, SmartThings Pet Care, and Family Care, has previously stated that microphones can detect dog barks and infant cries to send notifications. This functionality requires activation within SmartThings and is supported on select high-end models with sound-sensing capabilities.
These two perspectives highlight a divergence: Samsung emphasizes pet and family care features, while the researcher questions the scope of consent, transparency, and the collection of data for advertising. Another post mentioned code for facial bounding boxes, but current evidence only suggests the ability to outline faces, not to identify individuals.
OpenAI Ban Amidst Research
On the other side of this controversy is how AI platforms handle researchers. On the evening of October 23rd, @log4j posted that he was banned from OpenAI for his reverse-engineering activities, citing a loss of approximately $500 in credits and “banked resets,” and included a screenshot of an email labeled “Cyber Abuse.”
His account of the situation is that he never used OpenAI for vulnerability research or attacks, only for local reverse engineering, which he claims triggered the ban. OpenAI has not publicly commented on this specific suspension, and Samsung has yet to respond publicly to his detailed accusations.
What is currently clear is that this series of posts has brought “on-device AI and advertising telemetry” into the spotlight, attracting media attention. However, the full technical report remains unpublished. The next steps will depend on the details of the report, whether official responses are forthcoming, and if the alleged behaviors can be independently replicated on unmodified Samsung TV models.
Leave a Reply