Viruses and their hosts have been engaged in a continuous evolutionary arms race for billions of years. To replicate, viruses hijack the host translational machinery and redirect the cell's metabolic resources for their own benefit. In response, cellular organisms rely on diverse antiviral mechanisms that detect and counteract viral infections. Many of these antagonistic strategies are mediated by protein–protein interactions (PPIs), through which host factors target viral components for degradation or viral factors disrupt the normal course of cellular processes. Recent advances in protein structure prediction, particularly with AlphaFold2-Multimer and AlphaFold3, have enabled large-scale, structure-guided exploration of potential PPIs at the virus-host interface. Building on these advances, we used AlphaFold3 to model thousands of putative interactions between 30 human antiviral factors (including members of the OAS, IFITM, and TRIM protein families) and proteins from 10 medically important viruses, including HIV, Chikungunya, and Zika. Our analysis highlighted both the strengths and current limitations of AlphaFold-based PPI modelling. Although the overall discovery rate of PPIs is modest, it can be enhanced by incorporating multiple-seed sampling and domain-level modelling. Meanwhile, the prediction reliability can potentially be improved by combining ipTM with ipSAE confidence metrics to better distinguish true positive hits from modelling artefacts. Overall, this large-scale modelling approach has strong potential for systematically exploring virus–host PPIs and prioritising promising interaction candidates for experimental validation.