The relative contribution and physical drivers of internal variability in recent Arctic warming remain as open questions leaving up for debate whether global climate models used for climate projection lack sufficient sensitivity in the Arctic to climate forcing. Here, through an analysis of large ensembles of fully coupled climate model simulations with historical radiative forcing, we present an important internal mechanism arising from low-frequency Arctic atmospheric variability in models that can cause substantial Arctic warming in addition of that due to anthropogenic forcing. This simulated internal variability shows a strong similarity to the observed Arctic atmospheric change in the past 37 years. Through a “fingerprint” pattern matching method, we estimate that this internal variability contributes to about 40 to 50% of observed multi-decadal decline in Arctic sea ice and temperature warming trends. Our study also suggests that global climate models may not actually underestimate climate sensitivities in the Arctic but have trouble fully replicating an observed linkage between the Arctic and the lower latitudes in recent decades. Further improvements in simulating the observed Arctic-global linkage are thus necessary before the Arctic’s sensitivity to global warming in models can be quantified with confidence.