Predicting the changes in macroscopic properties of steels subjected to irradiation is key for the improved and long term safety of nuclear power plants. Because of the inherent complexity of the processes involved, this is a long-lasting challenge for nuclear materials science. Experience accumulated during the last few decades shows that nuclear reactor pressure vessel (RPV) steels harden and embrittle because of the formation of nano-sized clusters of their minor alloying elements. In this work, we combine a variety of experimental studies, mainly based on atom probe tomography (APT) and partly on small-angle neutron scattering (SANS), with recent theoretical advances, to rationalise the processes leading to the formation of embrittling solute nano-clusters. We show that the mechanism of solute cluster formation is dominated by the interplay between radiation-defects, their clusters, and solute atoms. This deviates from classical nucleation and growth processes that often drive, for example, phase separation under thermal ageing. In particular, the combination of solute dragging by point defects towards sinks, and the affinity between solutes and point defect clusters, turn out to be the key mechanisms for the formation of solute clusters, that do not need to reach a critical size to become metastable, while not being necessarily thermodynamically stable. The radiation-induced component to hardening and embrittlement of steels is therefore dominant and the fundamental mechanisms that have been identified provide a generalised theoretical framework to understand radiation effects in metallic alloys. To demonstrate this, we develop a model based on the latest theoretical findings that describes correctly the kinetics of formation of solute-rich clusters in a variety of not only RPV steels, but also model alloys and ferritic-martensitic steels, subjected to different irradiation conditions. The model accurately explains the APT and SANS evidence for all these materials. This opens the way to the creation of a physically-based computational tool that provides reliable assessments of radiation embrittlement in steels as a function of composition, temperature, and neutron dose and dose-rate.