BACE1 (β-secretase) and α-secretase cleave the Alzheimer's amyloid β protein (Aβ) precursor (APP) to C-terminal fragments of 99 aa (CTFβ) and 83 aa (CTFα), respectively, which are further cleaved by γ-secretase to eventually secrete Aβ and Aα (a.k.a. P3) that terminate predominantly at residues 40 and 42. A number of γ-secretase inhibitors (GSIs), such as N-[N-(3,5-Difluorophenacetyl-L-alanyl)]-S-phenylglycine t-butyl ester (DAPT), have been developed with the goal of reducing Aβ to treat Alzheimer's disease (AD). Although most studies show that DAPT inhibits Aβ in a dose-dependent manner several studies have also detected a biphasic effect with an unexpected increase at low doses of DAPT in cell cultures, animal models and clinical trials. In this article, we confirm the increase in Aβ40 and Aβ42 in SH-SY5Y human neuroblastoma cells treated with low doses of DAPT and identify one of the mechanisms for this paradox. We studied the pathway by first demonstrating that stimulation of Aβ, a product of γ-secretase, was accompanied by a parallel increase of its substrate CTFβ, thereby demonstrating that the inhibitor was not anomalously stimulating enzyme activity at low levels. Secondly, we have demonstrated that inhibition of an Aβ degrading activity, endothelin converting enzyme (ECE), yielded more Aβ, but abolished the DAPT-induced stimulation. Finally, we have demonstrated that Aα, which is generated in the secretory pathway before endocytosis, is not subject to the DAPT-mediated stimulation. We therefore conclude that impairment of γ-secretase can paradoxically increase Aβ by transiently skirting Aβ degradation in the endosome. This study adds to the growing body of literature suggesting that preserving γ-secretase activity, rather than inhibiting it, is important for prevention of neurodegeneration.