Branching regulatory biographies of GPT-4 and DeepSeek-R1: A BOAP-informed analysis of regulatory reachability in the European Union, the United States and China

Abstract

Using a documentary biography informed by Biographies of Artefacts and Practices (BOAP), this article traces how branches of GPT-4 and DeepSeek-R1 became reachable through regulatory pathways in the European Union, United States and China. Drawing on 132 documentary observations, it reconstructs five cross-site configuration moments from 2022 to 2025, supplemented by a bounded mid-2026 status check. The matched comparison covers hosted and API branches of both releases, while DeepSeek-R1's repository, derivatives and public serving routes extend the analysis to redistributed open-weight branches. Governance initially operated through inherited platform, algorithmic-service, data and compute controls before more explicit provider-, model- and output-oriented pathways emerged. Provider-controlled services preserved identifiable actors and control points, whereas open-weight redistribution allowed model lineage to persist across repositories, modifiers, hosts, deployers and platforms beyond the original developer's continuing control. Organisational responses further reconfigured later control points. The article develops generative AI in context as a BOAP-informed extension for model lineages whose technical continuity persists while distribution relationships and organisational control diverge. Branching regulatory biography captures this divergence, while regulatory reachability identifies which instruments can address which actors through which control points for a particular branch and time. Distributed-model governance therefore requires differentiated responsibilities and portable records, while treating documentation as support for, rather than a substitute for, accountable legal actors.