Extreme mass ratio inspirals (EMRIs) occur when stellar-mass compact objects begin a gravitational wave (GW) driven inspiral into massive black holes. EMRI waveforms can precisely map the surrounding spacetime, making them a key target for future space-based GW interferometers such as LISA, but their event rates and parameters are massively uncertain. One of the largest uncertainties is the ratio of true EMRIs (which spend at least thousands of orbits in the LISA band) and direct plunges, which are in-band for at most a handful of orbits and are not detectable in practice. In this Letter, we show that the traditional dichotomy between EMRIs and plunges-EMRIs originate from small semimajor axes, plunges from large-does not hold for intermediate-mass black holes with masses M_{•}≲10^{5}M_{⊙}. In this low-mass regime, a plunge always has an O(1) probability of failing and transitioning into a novel "cliffhanger" EMRI. Cliffhanger EMRIs are more easily produced for larger stellar-mass compact objects, and are less likely for smaller ones. This new EMRI production channel can dominate volumetric EMRI rates n[over ˙]_{EMRI} if intermediate-mass black holes are common in dwarf galactic nuclei, potentially increasing n[over ˙]_{EMRI} by an order of magnitude.