Superluminous supernovae (SLSNe) are a rare class of transients with peak luminosities 10-100 times greater than those of standard core-collapse supernovae (SNe). The mechanisms powering their extreme brightness remain debated, with circumstellar medium (CSM) interaction, or energy injection from a central engine like a magnetar wind nebula being the most plausible scenarios. To further constrain the underlying mechanism, we carried out a systematic search for GeV gamma-ray emission using the Fermi-LAT telescope from a sample of nearby hydrogen-poor (Type I) and hydrogen-rich (Type II) SLSNe over the past 16 years. Among the sample, only SN 2017egm shows significant gamma-ray emission, with likelihood test statistic (TS) values of 26-33 (i.e., >5$\sigma$) depending on the adopted time window. The signal arises between 50 and 160 days after explosion and is well described by a power-law spectrum with index $\Gamma=2.17 \pm 0.23$. The emission is consistent both in terms of its light curve and its spectrum, with predictions from magnetar models requiring either low nebular magnetization or faster spin-down than dipole losses. The CSM shell interaction scenario can reproduce the observed flux level but not the observed timing of the gamma-ray signal. In addition, the observed ratio, $L_{\gamma}/L_{opt} \sim 1$, is inconsistent with theoretical expectations and not in line with ratio measurements in other interacting CSM-dominated objects (e.g., novae or SNe) where this ratio is less than $10^{-2}$. Our study strongly suggests that a central engine like a magnetar plays a key role in this SLSN and could explain the bulk of the optical and gamma-ray light curves properties. Finally, simulations of 50 hours of CTAO observations indicate that a SN 2017egm-like event would be detectable up to 140 Mpc in the magnetar model but not in the CSM model due to strong gamma-gamma absorption.