Extension: Sovereign Agents: Towards Infrastructural Sovereignty and Diffused Accountability in Decentralized AI
Cryptographic self-custody can secure operational sovereignty without supplying institutional accountability. Hu and Rong show that decentralized agents derive persistence and resistance to override from cryptographic keys, trusted execution environments, and distributed protocols. The same layered architecture diffuses responsibility across developers, infrastructure providers, protocol governance, and economic participants. Their analysis independently extends Kaal's distinction: control over custody identifies where assets sit and who can reach them, but does not identify who acted under authority or who answers for a harmful outcome. The source is a 2026 arXiv preprint submitted to FAccT, not a peer-reviewed final article. It analyzes decentralized agents and infrastructural hardness. It does not establish that every local runtime lacks accountability. The evidence supports a narrower institutional conclusion: sovereign custody must be joined to identifiable actors, enforceable duties, and remedial consequences.
institutional-designgovernance-designai-and-agentsdata-sovereigntyaccountabilityevidence-provenance