Health-conscious battery management systems (BMS), considering only the surface temperature is insufficient for automotive lithium-ion batteries (LIBs). Experimental studies have revealed differences between cylindrical LIBs surface and core temperatures under dynamic current loading. Therefore, battery management considers only the sample surface temperature, missing this vital information. This results in significant delays in identifying thermal events inside the cell, leading to accelerated battery degradation and thermal runaway. To obtain accurate information on the core temperature, this paper introduces a Kolmogorov-Arnold Network (KAN), and a long shortterm memory (LSTM) network without necessity of using the surface temperature as feedback to the neural network. Experimental validation demonstrated an error of 0.5 • C and computational cost of 2.9 ms to 3.2 ms while estimating the core temperature for the KAN. The proposed technique estimates surface and core temperature, thus eliminating the need for surface temperature sensor feedback while estimating the core temperature. The proposed KAN-based estimation is not only adaptive to changes in operating conditions to maintain accuracy throughout the battery life, but it also keeps computational costs within 2.9 ms to 3.2 ms, acceptable limits for onboard BMS. Thus, it can be considered qualified for use in onboard BMS and cloudenabled digital-twin-based BMS.