Language Models as a Challenge for Business Ethics – A Partially Open-Source Approach

Abstract

Abstract Large language models have become central infrastructures of contemporary digital economies while raising persistent ethical concerns regarding linguistic inequality, opacity, data governance, and the concentration of technological power. Much of the current debate on AI ethics focuses on normative principles such as fairness, transparency, and accountability. While these principles remain essential, they often do not sufficiently explain why ethically problematic outcomes persist under competitive market conditions. This paper addresses that gap by applying Karl Homann’s institutional economic ethics to the governance of large language models. From this perspective, ethical deficits in AI development are not merely the result of individual failures of responsibility, but are also shaped by institutional incentive structures that reward speed, scale, proprietary control, and strategic secrecy. The paper analyses three central ethical challenges in LLM development: linguistic and cultural asymmetries, transparency and accountability deficits, and contested practices of data governance and intellectual property. It then argues that the familiar opposition between open and closed AI systems is conceptually and institutionally inadequate. In response, the paper develops the concept of partially open AI governance, understood as a differentiated arrangement of access, disclosure, and oversight across distinct layers of AI systems. Such an approach offers a more realistic way of aligning innovation incentives with ethical and public objectives in the governance of large language models.