SPIES Research Lab’s study, led by Kalyan Nakka, Jimmy Dani and Nitesh Saxena, has been cited in a major South Korean news coverage, featuring a significant AI security incident of LG Uplus’s popular AI calling app “ixi-O”. Our study on the trust and ethics gap in Small Language Models (SLMs) applied to on-device AI, has revealed that SLMs have “much lower reliability than server-based SLMs” and pose “greater risk of leaking personal information”. These findings proved prescient, when “ixi-O” app, promoted as strong on security due to on-device AI, has experienced a data breach affecting dozens of users in South Korea, emphasizing our study’s privacy concerns in on-device AI systems and their real-world impact.
Link: https://biz.chosun.com/en/en-it/2025/12/09/LYNFHR3PKJAVDHXP4LWOBIHHXQ/
Read More: arXiv
