Observed fingerprint of global warming exposes model biases in regional climate attribution.

Disentangling externally forced signals from internal variability in observed surface air temperature (SAT) is essential for reliable regional climate detection and attribution. Although climate model simulated patterns broadly align with observed global warming, persistent regional discrepancies, compounded by the limitation of a single observational realization, impede robust regional attribution. We identify the observation-based SAT externally forced signals and internal variability, which allows us to compare the observed SAT patterns with the simulated counterparts from seven CMIP6 single-model initial-condition large ensembles. Perfect-model tests confirm robust large-scale partitioning, while indicating reduced fidelity where internal variability dominates. Our results suggest that models tend to overestimate emergence timescales in externally forced regions but underestimate them in internal variability-dominated regions. In several internal variability-dominated regions, the forced signal remains undetectable in observations. While Arctic warming is increasingly dominated by external forcing, internal variability has slowed its rate in recent decades. In internal variability-dominated regions, we identify modest yet consistent externally forced contributions, with observed cooling trends and sign flips primarily driven by internal variability. These findings help to resolve long-standing regional attribution puzzles, highlight critical model limitations, and carry fundamental implications for regional climate projections, attribution, and jurisprudence.