The phenomenological interplay in the segregation of phosphorus (P) and transition metal (TM) elements at grain boundaries (GBs) in steels has long been suspected to be the main contributor to temper embrittlement. However, many of the details remain unclear. Here, we investigate the segregation, co-segregation and cohesion effects of TMs (Co, Cr, Cu, Mn, Mo, Ni, Nb, Ti, V, W) along with P in various ferritic iron (-Fe) GBs utilising density functional theory and simulations of kinetics. Our findings demonstrate that P is unlikely to cause intergranular fracture via weakened interfacial bonding when segregated by itself. Nevertheless, the stronger segregation binding of P compared to TMs can explain the ubiquitous presence of P segregated at GBs. We find that most P-TM interactions at ferritic GBs are repulsive and differ significantly from the corresponding interactions in the bulk. Due to the repulsive interactions and strong segregation binding of P, the enrichment of P over time at GBs leads to the depletion of prior-segregated cohesion-enhancing solutes at general GBs. Additionally, certain P-TM co-segregation combinations that are cohesion-lowering are energetically favoured at such GBs. We posit these mechanisms act in tandem as critical causalities of P-induced temper embrittlement in alloyed steels. Finally, we reveal a contradiction in the predicted cohesion effects of segregated P calculated in the Rice–Thomson–Wang theory of interfacial embrittlement compared to that as assessed by chemical bonding strength, calculated in the DDEC6 bond-order framework. These findings have important implications for GB engineering for interfacial cohesion.