Decentralized AI Creates Ungovernable Systems, New Paper Argues

A recent study published on arXiv (ID 2605.24538) claims that decentralized AI (DeAI) systems cannot be effectively governed within current regulatory structures. The researchers examine DeAI through a six-layer framework—model, training, compute, harness, identity, and ownership—and demonstrate that partial decentralization leads to a 'governance vacuum.' This vacuum consists of two distinct gaps: an accountability gap, where no clear responsible party exists, and an incapacitation gap, where even if a party is identified, they are unable to modify the operational system. The authors argue that these shortcomings undermine the foundational assumptions of existing AI governance, which depends on identifiable developers, deployers, or operators.