The　Web　3.0　era　is　reshaping　the　way　we　share　data,　and　online　social　networks　play　an　essential　role　in　this　process.　In　online　social　networks　(OSN),　the　interaction　between　users　is　closely　related　to　data　security.　With　the　widespread　use　of　OSN,　the　harm　caused　by　fake　news　has　penetrated　many　corners　of　society.　Most　existing　graph　machine　learning　works　for　fake　news　detection　focus　only　on　the　news　propagation　path　or　news　content　itself;　they　ignore　the　trust　relationships　between　users　and　the　ease　of　attack　of　the　graph　formed　by　the　propagation　path.　The　underlying　trust　factors　among　users　can　be　revealed　by　their　endogenous　preferences　that　help　indicate　the　extent　to　which　others　are　expected　to　perform　particular　actions.　Moreover,　joint　trainable　news　propagation　paths　and　social　trust　can　improve　the　robustness　of　graph　network　models　and　slow　the　accumulation　of　errors　caused　by　fraudulent　messaging.　However,　these　works　are　somewhat　limited　in　fake　news　detection.　This　paper　proposes　novel　robust　trust　evaluation　architecture　for　fake　news　detection,　RTrust,　which　improves　the　performance　of　fake　news　detection　by　incorporating　trust　propagation　and　robustness.　Comparative　results　from　the　latest　baseline　on　two　real-world　datasets　demonstrate　the　advantages　of　RTrust　in　detecting　fake　news　and　its　robustness.　©　The　Author(s),　under　exclusive　licence　to　Springer　Science+Business　Media,　LLC,　part　of　Springer　Nature　2024.