Evolutionary turnover in non-coding regions has driven phenotypic divergence during past speciation events and continues to facilitate environmental adaptation through variants. We used a deep learning model to identify the substrates of regulatory turnover using genome wide mutations mimicking three evolutionary pathways: recent history (human-chimp substitutions), modern population (human population variation), and mutational susceptibility (random mutations). We observed enhancer turnover in approximately 6% of the whole genome, with more than 80% of the novel activity arising from repurposing of enhancers between cell-types. Frequency of turnover in a cell-type is remarkably similar across the three pathways, despite only ~19% overlap in the source regions. The enhancers predisposed to turnover display reduced evolutionary constraints and are depleted in GWAS variants. Most of these occur within 100kb of a gene, with the highest turnover occurring near neurodevelopmental genes including CNTNAP2, NPAS3, and AUTS2. Flanking enhancers of these genes undergo high turnover irrespective of the mutational model pathway, suggesting a high plasticity in neurocognitive evolution. Based on susceptibility to random mutations, these enhancers were identified as vulnerable by nature and feature a higher abundance of cell type-specific transcription factor binding sites (TFBSs). Our findings suggest that enhancer repurposing within vulnerable loci drives regulatory innovation while keeping the core regulatory networks intact.