Paper Number

ICIS2026-2116

Paper Type

Short

Abstract

Governments are developing sovereign large language models (LLMs) to align generative AI with public-sector values and legal frameworks. Yet research still treats LLMs as stable artifacts, overlooking their continuous reconfiguration. This paper examines Sweden's SVEA initiative as public-sector AI-in-the-making, drawing on documents, interviews, and focus groups across two development phases. We identify four coexisting versions of SVEA—a technical model, a governance object, a political project, and a workplace tool—and three frictions through which they develop: an annotation knot binding retrieval accuracy to exhausting labor; a sovereignty gap in which political ambition and technical capability alike lack a legal form; and a localization cascade in which political boundaries become hardware contracts and then workplace capabilities. We contribute a four-version framework for public-sector AI infrastructures and show that national-scale versioning lacks an institutional stabilizer: versions are produced by institutions that cannot direct one another and must be held together jurisdictionally.

Comments

18_IoT_publicIS_infrastructures

Share

COinS
 
Dec 14th, 12:00 AM

SVEA-in-the-Making: A Sociomaterial Genealogy of Public-Sector LLM Development in Sweden

Governments are developing sovereign large language models (LLMs) to align generative AI with public-sector values and legal frameworks. Yet research still treats LLMs as stable artifacts, overlooking their continuous reconfiguration. This paper examines Sweden's SVEA initiative as public-sector AI-in-the-making, drawing on documents, interviews, and focus groups across two development phases. We identify four coexisting versions of SVEA—a technical model, a governance object, a political project, and a workplace tool—and three frictions through which they develop: an annotation knot binding retrieval accuracy to exhausting labor; a sovereignty gap in which political ambition and technical capability alike lack a legal form; and a localization cascade in which political boundaries become hardware contracts and then workplace capabilities. We contribute a four-version framework for public-sector AI infrastructures and show that national-scale versioning lacks an institutional stabilizer: versions are produced by institutions that cannot direct one another and must be held together jurisdictionally.

When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.