ABSTRACT What happens when artificial intelligence (AI) produces authoritative but false information in public decision‐making? This article examines an AI‐induced crisis in a Norwegian municipality where bureaucrats relied on ChatGPT‐generated content in a report to determine whether eight schools were to be permanently closed. The content was presented as legitimate research but was in fact hallucination by the AI technology, leading to an acute crisis demanding a response from different public actors. We find that the involved actors managed reputation and accountability in response to the AI‐induced crisis in quite inconsistent and contradictory ways, shaped by their institutional roles and differing interpretations of accountability. Building on the empirical findings, we introduce the concept of the double black box to illustrate how overlapping technological opacity and governance opacity can complicate transparency and obscure accountability for public decisions.
Lund‐Tønnesen et al. (Tue,) studied this question.