Nuclear reactor cores achieve sustained fission chain reactions through the so-called "critical state"-a subtle equilibrium between their material properties and their geometries. Observed at macroscopic scales during operations, the resulting stationary neutron field is tainted by a noise term that hinders various fluctuations occurring at smaller scales. These fluctuations are either of a stochastic nature (whenever the core is operated at low power) or related to various perturbations and vibrations within the core, even operated in its power regime. For reasons that are only partially understood using linear noise theory, incidental events have been reported, characterized by an increase of the power noise. Such events of power noise growth, sometimes up to seemingly unbounded levels, have already led in the past to voluntary scramming of reactors. In this paper, we will use a statistical field theory of critical processes to model the effects of neutron power noise. We will show that the evolution of the neutron field in a reactor is intimately connected to the dynamic of surface growths given by the Kardar-Parisi-Zhang equation. Recent numerical results emerging from renormalization-group approaches will be used to calculate a threshold in the amplitude of the reactor noise above which the core enters a new criticality state, and to estimate the critical exponents characterizing this phase transition to rough neutron fields. The theoretical model of nonlinear noise built in this paper from ab initio statistical mechanics principles will be correlated and compared to data of misunderstood reactor noise levels and reactor instabilities and will be shown to provide both qualitative and quantitative insights into this long-standing issue of reactor physics.