Risk assessments of air pollution impacts on human health and ecosystems would ideally consider a broad set of climate and emission scenarios and the role of natural internal climate variability within a single scenario. We analyze initial condition chemistry-climate ensembles to gauge the significance of greenhouse-gas-induced air pollution changes relative to internal climate variability, and response differences in two models. To quantify the effects of climate change on the frequency and duration of summertime regional-scale pollution episodes over the Eastern United States (EUS), we apply an Empirical Orthogonal Function (EOF) analysis to a 3-member GFDL-CM3 ensemble with prognostic ozone and aerosols and a 12-member NCAR-CESM1 ensemble with prognostic aerosols under a 21^st century RCP8.5 scenario with air pollutant emissions frozen in 2005. Correlations between GFDL-CM3 principal components for ozone, PM_2.5 and temperature represent spatiotemporal relationships discerned previously from observational analysis. Over the Northeast region, both models simulate summertime surface temperature increases of over 5 °C from 2006-2025 to 2081-2100 and PM_2.5 of up to 1-4 μg m^-3. The ensemble average decadal incidence of upper quartile Northeast PM_2.5 events lasting at least five days doubles in GFDL-CM3 and increases >50% in NCAR-CESM1. In other EUS regions, inter-model differences in PM_2.5 responses to climate change cannot be explained by internal climate variability. Our EOF-based approach anticipates future opportunities to data-mine initial condition chemistry-climate model ensembles for probabilistic assessments of changing frequency and duration of regional-scale pollution and heat events while obviating the need to bias-correct concentration-based thresholds separately in individual models.