Additive noise is known to affect the stability of nonlinear systems. To understand better the role of additive noise in neural systems, we investigate the impact of additive noise on a random neural network of excitatory and inhibitory neurons. Here we hypothesize that the noise originates from the ascending reticular activating system (ARAS). Coherence resonance in the γ-frequency range emerges for intermediate noise levels while the network exhibits non-coherent activity at low and high noise levels. The analytical study of a corresponding mean-field model system explains the resonance effect by a noise-induced phase transition via a saddle-node bifurcation. An analytical study of the linear meanfield systems response to additive noise reveals that the coherent state exhibits a quasi-cycle in the γ−frequency range whose spectral properties are tuned by the additive noise. To illustrate the importance of the work, we show that the quasi-cycle explains γ−enhancement under impact of the anaesthetics ketamine and propofol as a destabilizing effect of the coherent state.