LG Uplus to launch AI-driven anti-phishing technology

LG Uplus Senior Vice President Jeon Byoung-ki speaks during a press event in Seoul, Thursday. Courtesy of LG Uplus

LG Uplus is introducing a new artificial intelligence (AI)-powered security feature for its ixi-O AI agent, in response to the growing number of cases of voice phishing and fraud crimes using AI-driven deepfake technologies.

According to the company, the new anti-deepvoice technology will be integrated into ixi-O on Monday. The technology was unveiled at this year’s MWC tech trade show in March.

The new technology, which the company is commercializing for the first time globally as an on-device AI, detects voices made or manipulated by AI during a call and alerts the user within five seconds. It aims to help identify and prevent voice phishing crimes using AI-generated voices to impersonate acquaintances or stage fake hostage situations to extort money.

“AI-generated voices may not be noticeable to our ear, but they often contain abnormal high-frequency noise or unnatural variations in pronunciations that differ from real humans. By using AI, we are able to detect these subtle differences that people typically can’t perceive,” Lee Seok-young, director of mobile services at LG Uplus, said during a press event in Seoul.

To develop the technology, LG Uplus trained its AI engine with about 3,000 hours of call data, equivalent to about 2 million calls. It uses the company’s own technologies such as voice activity detection to identify the actual voice segment during calls, speech-to-text to convert the detected voice into text and anti-spoofing to analyze whether the voice is manipulated.

While developing the technology, LG focused on making the technology lightweight for ixi-O to operate as an on-device AI without storing user information on the telecom’s server, which reduces the risks of leaking users’ personal data.

A model demonstrates ixi-O's new anti-deepvoice feature. Courtesy of LG Uplus

The company is expanding the feature to analyze calling patterns from phone numbers previously reported for phishing to predict and flag high-risk calls, as well as automated answering of those calls for users.

It has also partnered with the National Forensic Service to develop a system to detect criminal voices, which will compare the caller’s voice in real time to the registered voiceprints of phishing criminals and send an alert if a high similarity is detected.

Additionally, LG Uplus has developed anti-deepfake technology that can identify AI-generated videos of a person, which is yet to be commercialized. The technology analyzes videos or images to detect unnatural traces left behind in synthetic videos, such as pixel-level textures and inconsistencies or disruptions in frame continuity.

“By offering our differentiated security technology, we aim to establish ourselves as the telecom company that is most committed to security and holds the safest AI-powered technology,” said Jeon Byoung-ki, senior vice president and the head of the AI and data tech group.

“We will deliver real, tangible value so that AI can make our customers’ lives safer and help realize the ‘brighter world’ LG Uplus envisions.”