Historical failures and epistemic accidents: NASA’s Challenger disaster and the organizational risks of Generative AI

Purpose This paper aims to develop a recursive model of how organizational pressures for simplification drive the repeated adoption of technologies that undermine expert judgment. By examining the epistemic mechanisms that contributed to NASA’s Space Shuttle Challenger disaster and tracing their structural recurrence in the organizational adoption of Generative Artificial Intelligence, the study reveals systemic risks in technology-mediated knowledge systems. Design/methodology/approach The study uses a historically grounded conceptual argument using abductive reasoning. Synthesizing the Rogers Commission Report, Vaughan’s organizational ethnography and Tufte’s analysis of representational compression with theories of normal accidents (Perrow), organizational epistemology (Weick; Spender) and epistemic machinery (Kaplan), the analysis iterates between historical evidence and theoretical frameworks to identify recurring mechanisms of epistemic failure. Findings The study identifies three mechanisms through which Generative Artificial Intelligence may reproduce epistemic risks analogous to those that led to the Challenger disaster: token-based epistemic filtering, algorithmic epistemic substitution and interactive complexity with tight coupling. These mechanisms combine within a recursive model that shows how institutional simplification pressures create self-perpetuating cycles of epistemic vulnerability. Research limitations/implications As a conceptual study, the model requires empirical validation. Future research should conduct comparative case studies and test whether epistemic resilience practices reduce failures in organizational settings. Practical implications Organizations should implement epistemic resilience practices, including mandatory expert verification in high-stakes contexts, transparency requirements for processing boundaries, epistemic red teams, hybrid workflows and audit trails. Social implications The adoption of Generative Artificial Intelligence risks shifting epistemic authority from accountable professionals to opaque algorithmic systems, creating asymmetric power dynamics and potential harm in high-stakes domains. Originality/value This study extends normal accident theory to knowledge systems, introduces algorithmic epistemic substitution and token-based epistemic filtering as mechanisms of organizational risk and theorizes epistemic resilience practices translating high-reliability organizing principles into epistemic safety.