We demonstrate a machine learning attack resilient true random number generator (TRNG) based on MoS2 Fe-FET. The probabilistic nature of the ferroelectric-accelerated charge trapping/de-trapping mechanism is exploited as the entropy source to achieve (i) a significantly reduced operation voltage to trigger the stochastic process by introducing a ferroelectric HZO layer with a writing energy (Ewrite) of ~0.7 pJ/bit; (ii) a physically unclonable Fe-TRNG with a near-ideal entropy of 0.99; (iii) a designed self-correction circuitry that further improves its randomness with no observable degradation after 3×106 endurance cycles. Additionally, we demonstrate the random quick response (QR) passwords as a protection layer for enhancing the hardware security in IoT. A seed throughput of 1250 bit/s and an estimated circuit power consumption of ~0.37 mW are featured in the Fe-TRNG array.