Data Exhaust Governance: AI’s Demand for Data and the State’s Expanding Information Capacity

AI adoption in government is reshaping governance architectures in ways that extend far beyond automation or decision support. This paper argues that contemporary public-sector AI systems generate a form of data-dependent governance, wherein algorithmic requirements reorder how states define, collect, validate, and integrate information across administrative domains. Through cases from India, the European Union, and the United Kingdom, the paper demonstrates that AI’s most significant institutional effects arise not from model outputs but from the upstream political and organisational responses to model fragility, data gaps, and epistemic opacity. In India, state-funded prediction systems for health insurance fraud detection, crop-loss assessment, and welfare-benefit targeting routinely expose inconsistencies in district-level reporting, incomplete private-sector data, and heterogeneous coding practices. Rather than abandoning these systems, governments have responded by expanding mandatory reporting obligations, enforcing uniform coding standards, and accelerating cross-scheme data linkage. These changes are not neutral quality improvements; they consolidate particular forms of state legibility while increasing the administrative burden on low-capacity regions, reinforcing existing asymmetries in state-citizen visibility. A similar pattern emerges in the EU’s deployment of AI-enabled risk scoring in border management and labour inspection. Model performance audits conducted under the oversight of national supervisory authorities have repeatedly concluded that predictive accuracy is constrained by insufficiently harmonised member-state datasets. In response, several administrations have advanced proposals for deeper interoperability between migration, employment, and social security registries—changes that effectively reconfigure the information-sharing architecture of the EU before any formal political debate about the governance implications of such integration. In the UK, the National Audit Office’s reviews of algorithmic systems in policing and welfare have highlighted that the most significant governance problems stem from the opacity of data provenance and undocumented transformation pipelines. These findings have triggered internal reorganisations: the creation of dedicated data engineering units, new mandatory metadata standards, and the consolidation of dispersed data custodianship under fewer, more centralised bodies. As with the EU, these institutional redesigns were justified as technical necessities, yet they materially shift power, authority, and accountability relationships within the state. This paper makes three contributions. First, it develops the concept of data-dependent governance to explain how AI-induced information requirements reshape bureaucratic coordination, administrative hierarchies, and the locus of decision authority. Second, it shows how efforts to remedy upstream data weaknesses introduce new governance risks, including unmediated expansion of data infrastructures, weakened contestability of administrative knowledge, and reduced transparency of decision chains. Third, it argues that these shifts occur largely outside public deliberation, driven by the internal logic of algorithmic optimisation rather than explicit democratic choice. The paper concludes that AI in government is an architectural force one that restructures the informational and organisational foundations of governance and must be studied as such if institutional resilience and public values are to be preserved.