Climate change poses direct and indirect threats to public health, including exacerbating air pollution. However, how a warmer temperature deteriorates air quality, known as the "climate penalty" effect, remains highly uncertain in the United States, particularly under rapid reduction in anthropogenic emissions. Here we examined the sensitivity of surface-level fine particulate matter (PM2.5) and ozone (O3) to summer temperature anomalies in the contiguous US and their decadal changes using high-resolution datasets generated by machine learning models. Our findings demonstrate that, in the eastern US, efficient emission control strategies have significantly reduced the climate penalty effects on PM2.5 and O3, lowering the associated population exposure. In contrast, summer and annual PM2.5 in the western US became more sensitive to temperature, highlighting the urgent need for the management and mitigation of worsening wildfires. Our results have important implications for air quality management and risk assessments of future climate change.