Current highly pathogenic H5N1 avian influenza (HPAI H5N1) viruses in bovine and other mammals have been posing unprecedented risks to public health. It’s vital and urgent to assess the pandemic potential of the HPAI H5N1 virus, and the risk degree posed by the virus infection or the genome reassortment with human influenza A viruses (IAVs). An attentional deep learning framework here was constructed of Human Adaptive Influenza virus Reassortment using Attentional Networks based on Genome Embedding (HAIRANGE), to predict high-risk reassortment between avian and human IAVs. HAIRANGE embedded genomic contextual codons covering both RNA and protein information, biologically interpretable on viral adaptive codon contexts of IAVs, predicted accurately adaptive IAV genes and adaptive reassortment between avian and human IAVs on independent validation data sets of RNA polymerase-related genes. A high adaptive reassortment risk was predicted by HAIRANGE of the current bovine HPAI H5N1 viruses with human H3N2 IAVs, as has been in vitro validated with polymerase reporter assay. In summary, the present study provides an intelligent tool to predict high-risk IAV reassortment based on genome embedding. Current bovine HPAI H5N1 is posing high pandemic potential via possible genomic reassortment with human IAVs.