Abstract When the Australian Government’s automated debt recovery system issued hundreds of thousands of unlawful payment demands to welfare recipients, the error was not computational but interpretive: the algorithm embodied a reading of social security legislation that no court would have endorsed. This controversy—the Robodebt scandal—illustrates the stakes of Rules as Code, an emerging movement to translate legislation into machine-executable form. This article argues that such translation is not a neutral technical operation but a substantive interpretive undertaking, embedding contestable judgments about statutory meaning in systems applied automatically to an indefinite number of cases. Drawing on Hart’s account of open texture and Fuller’s principles of legality, the analysis demonstrates that the computational demand for determinacy conflicts with the deliberate flexibility of legal language. Evaluative standards such as ‘reasonable’ or ‘proportionate’ resist encoding without either defeating automation’s purpose or substituting quantitative thresholds for contextual judgment—though this article proposes that hybrid norms, bifurcating provisions into computationally tractable and human-assessed components, offer a promising middle path. On the contested question of legal status, this article argues that encoded rules should be recognized as a sui generis category of administrative instrument requiring its own regulatory framework, including mandatory interpretive documentation, formal certification of equivalence, and preserved citizen challenge rights. Comparative examination of emerging safeguards in Estonia, Kazakhstan, and France reveals that no jurisdiction has achieved a fully satisfactory accountability architecture. For courts, this article argues that judicial review should extend to the encoding itself—examining whether code faithfully implements statutory requirements, not merely whether code was correctly applied to particular facts.
Yermukhametova et al. (Thu,) studied this question.