Toward a Global Regime for Compute Governance - arXiv

Toward a Global Regime for Compute Governance: Building the Pause Button 
 
 
 
 
 
 
 

 
 

 
 
 
 
 
 \addbibresource 
 references.bib

 
 Toward a Global Regime for Compute Governance: Building the Pause Button

 
 
 Ananthi Al Ramiah
 
 Independent
 
 
 Raymond Koopmanschap
 
 Independent
 
 
 Josh Thorsteinson
 
 Independent
 
 
 Sadruddin Khan
 
 Independent
 
 
 Jim Zhou
 
 Independent
 
 
 Shafira Noh
 
 Independent
 
 
 Joep Meindertsma
 
 PauseAI
 
 
 Farhan Shafiq
 AI Safety Camp Project Lead
 Think Safe AI
 
 
 (June 25, 2025) 
 
 Abstract

 As AI capabilities rapidly advance, the risk of catastrophic harm from large-scale training runs is growing. Yet the compute infrastructure that enables such development remains largely unregulated. This paper proposes a concrete framework for a global Compute Pause Button : a governance system designed to prevent dangerously powerful AI systems from being trained by restricting access to computational resources. We identify three key intervention points—technical, traceability, and regulatory—and organize them within a Governance–Enforcement–Verification (GEV) framework to ensure rules are clear, violations are detectable, and compliance is independently verifiable. Technical mechanisms include tamper-proof FLOP caps, model locking, and offline licensing. Traceability tools track chips, components, and users across the compute supply chain. Regulatory mechanisms establish constraints through export controls, production caps, and licensing schemes. Unlike post-deployment oversight, this approach targets the material foundations of advanced AI development. Drawing from analogues ranging from nuclear non-proliferation to pandemic-era vaccine coordination, we demonstrate how compute can serve as a practical lever for global cooperation. While technical and political challenges remain, we argue that credible mechanisms already exist, and that the time to build this architecture is now, before the window for effective intervention closes.

 
 
 
 1 Executive Summary

 
 As AI capabilities accelerate, the risks associated with large-scale
training runs are growing rapidly. The most powerful AI systems —
those capable of triggering profound societal disruption or catastrophic
harm — require vast amounts of compute. These risks range from mass
disinformation, cyberattacks, and biological/chemical weapon
acceleration to economic destabilization, gradual loss of human agency,
and, in the extreme, existential risk through permanent loss of human
control. Yet current governance efforts overwhelmingly focus on AI use
and deployment, leaving the infrastructure that enables dangerous
training runs largely unregulated. According to a
recent estimate \parencite kokotajloAI20272025, at current rates
of progress, frontier labs could cross critical danger thresholds as
early as 2027–2028. Without credible mechanisms to pause or prevent
threshold-exceeding compute, we face the very real possibility of losing
control 

... (truncated, 98 KB total)