Year: 2026
Type: preprint
Abstract: As human–AI collaboration accelerates, the records of how decisions were reached (the reasoning, its history, its dissent and its status) are becoming infrastructure that people and institutions depend on. This paper asks who should own and govern such infrastructure. It argues that a system whose value depends on trust, neutrality and continuity across long periods is poorly placed in the hands of a single commercial owner, whose incentives can conflict with the function the system serves and whose lifetime is shorter than the continuity it promises. It also argues that commercial delivery is not the problem: capital, speed and product discipline matter, and a purely non-profit route has failures of its own. The paper proposes a stewardship model. A non-profit steward holds the rights, the standard and the open export format. An external ethics board with enforcement powers oversees the conditions, and a separate appeals panel hears challenges to its decisions. Commercial operators are licensees under binding conditions: no sale or mining of reasoning records, portability and the right to leave, audit rights, fair access, and no private ownership of institutional reasoning. The model is framed through four concerns: privacy, freedom, ownership and overreach. This is a statement of design intent and an argument, not an evaluation. The steward does not yet exist as a legal entity, no empirical claim is made about effects, and the paper sets out what would show that the model does not work.
Source: arXiv (Cornell University)
Authors: Chen, Jianan, Qin, Yufei, Lin, Yaosen
Subfield: Strategy and Management
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7214067803
Cites:
Cited by:
Related to: