Over the past thirty years, Arctic sea ice has rapidly declined. This has led to a greater interest in marine access to the region and a need for reliable ice predictions. However, the natural variability of sea ice is considerable and likely to change with continued reductions in sea ice. Additionally there is a need for improved understanding on the predictability characteristics of sea ice on interannual to decadal timescales. Here I present results that diagnose mechanisms giving rise to potential predictability in Arctic sea ice in Community Climate System Model (CCSM) integrations. This includes an analysis of statistical relationships from control and 20th-21st century integrations and how these change with long-term sea ice loss. It also includes analysis of perfect model studies designed to investigate the initial-value predictability of sea ice. These perfect model ensemble integrations are initialized with identical ice-ocean-terrestrial conditions allowing us to diagnose the potential predictability that resides with those initial conditions on seasonal to interannual timescales. Finally, a brief comparison to Antarctic sea ice predictability characteristics and mechanisms is provided.