We argue that in order to account for the muon anomalous magnetic moment $g-2$, dark matter and LHC data, non-universal gaugino masses $M_i$ at the high scale are required in the framework of the Minimal Supersymmetric Standard Model (MSSM). We also need a right-handed smuon $tildemu_R$ with a mass around 100 GeV, evading LHC searches due to the proximity of a neutralino $tilde{chi}^0_1$ several GeV lighter which allows successful dark matter. We discuss such a scenario in the framework of an $SU(5)$ Grand Unified Theory (GUT) combined with $A_4$ family symmetry, where the three $overline{5}$ representations form a single triplet of $A_4$ with a unified soft mass $m_F$, while the three $10$ representations are singlets of $A_4$ with independent soft masses $m_{T1}, m_{T2}, m_{T3}$. Although $m_{T2}$ (and hence $tildemu_R$) may be light, the muon $g-2$ and relic density also requires light $M_1simeq 250$ GeV, which is incompatible with universal gaugino masses due to LHC constraints on $M_2$ and $M_3$ arising from gaugino searches. After showing that universal gaugino masses $M_{1/2}$ at the GUT scale are excluded by gluino searches, we provide a series of benchmarks which show that while $M_{1}= M_{2} ll M_3$ is also excluded by chargino searches, $M_{1}< M_{2} ll M_3$ is currently allowed. Even this scenario is almost excluded by the tension between the muon $g-2$, relic density, Dark Matter direct detection and LHC data. The surviving parameter space is characterised by a higgsino mass $mu approx -300$ GeV, as required by the muon $g-2$. The LHC will be able to fully test this scenario with the upgraded luminosity via muon-dominated tri- and di-lepton signatures resulting from higgsino dominated $tilde{chi}^pm_1 , tilde{chi}^0_2$ and $tilde{chi}^+_1 , tilde{chi}^-_1$ production.