ABSTRACT Seismic risk assessment studies typically rely on fragility functions/curves to characterize the susceptibility of an asset or a class of assets to damage. These curves are essentially a summarization of responses evaluated on the basis of ground motions with record characteristics and intensities that may or may not be tied to a specific site and its hazard. As is typical with intermediate aggregations in a lengthy analysis chain, this operation achieves ease of use by condensing the behavior of each asset into one or more fragilities, at the cost of losing non‐negligible information. Specifically, the within‐event correlation in response among different assets is removed, eliminating the potential to distinguish a benign from an aggressive event. Whereas this may not be important for assessing damage and loss for a single asset or an ensemble of independent ones, it can become highly influential for an interconnected system of assets, such as critical energy infrastructure, and for compound (concomitant or cascading) hazards. Instead, disaggregating the fragility back into individual ground motion records and separately propagating the consequences to the overall system improves fidelity at the cost of complexity. To investigate the effect of conventional versus disaggregated fragilities, different simplified system topologies are employed, demonstrating cases where the loss of response correlation can lead to severe bias in the impact assessment at the system level.