LG Uplus: "Some Features Not Fully On-Device, Processed via Server... Improvements Planned"
"Call Summary Data Stored for Six Months With Customer Consent"
"All Possible AI Features to Be Processed On-Device in the Future"
LG Uplus has addressed security concerns regarding the exposure of call information on its artificial intelligence (AI) call assistant app "ixi-O" by publicly disclosing the scope of on-device AI application and its server storage structure. As the company has consistently emphasized its on-device foundation, it aims to correct any customer misunderstandings about how each function is processed and has announced plans to transition all possible features to on-device processing in the future.
On December 10, LG Uplus stated in a reference document, "On-device AI processes computations directly on the device, providing high security and real-time performance. However, due to current technological limitations, some features of ixi-O are operated through servers." The company added, "In light of this incident, we will clearly communicate the scope of on-device application and data storage structure."
According to LG Uplus, ixi-O uses on-device AI embedded in the device to handle speech-to-text (STT) conversion, the 'Visible Call' feature that displays conversation as subtitles during calls, real-time voice phishing detection, and spoofed voice detection. The company emphasized that the call audio itself is neither transmitted to the server nor stored for extended periods.
In contrast, the call summary feature sends the text converted on the device to the server for summarization, and the summary results are stored on the server in encrypted form for up to six months with customer consent. This measure is intended to allow users to access previous summaries even after changing smartphones or reinstalling the app. The original text used for summarization is immediately deleted after the summary is generated.
LG Uplus also presented its future direction to process all possible AI functions on-device. Internal testing is underway to integrate an on-device small language model (sLM) based on "EXAONE 3.5 2.4B," co-developed with LG AI Research and OptAI, into ixi-O. Once implemented, the call summary feature can also be transitioned to an on-device basis. The AI voice search function, scheduled for introduction in the first half of next year, will further enhance security by transmitting only the content after the "Hey ixi" wake word to the server, without storing any content prior to the wake word.
Hot Picks Today
"Stocks Are Not Taxed, but Annual Crypto Gains Over 2.5 Million Won to Be Taxed Next Year... Investors Push Back"
- "Not Jealous of Winning the Lottery"... Entire Village Stunned as 200 Million Won Jackpot of Wild Ginseng Cluster Discovered at Jirisan
- Trump Holds Off on Iran Strike for "Two to Three Days"...Renews Pressure for Nuclear Abandonment (Comprehensive)
- "Looks Even More Like Him in Person": Albino Water Buffalo with Golden Hair and Pink Skin Nicknamed 'Trump'
- "Even With a 90 Million Won Salary and Bonuses, It Doesn’t Feel Like Much"... A Latecomer Rookie Who Beat 70 to 1 Odds [Scientists Are Disappearing] ③
LG Uplus stated, "Taking this call information leak as an opportunity, we will thoroughly review and improve all ixi-O processes, and strive to provide safer and more convenient AI services."
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.