An International Agreement to Prevent the Premature Creation of Artificial Superintelligence

TL;DR

The paper tackles the risk of premature artificial superintelligence by proposing an international, verifiable halt that postpones ASI until robust alignment and mitigation strategies exist. Its core approach combines FLOP-based training limits, comprehensive AI chip tracking, and targeted restricted research under a US–China-led coalition and a Coalition Technical Body, complemented by challenge inspections and whistleblower protections. It provides a concrete draft agreement with enforcement and verification mechanisms, plus a staged implementation framework to build political support while gradually expanding oversight. The work highlights the global governance implications of high-stakes AI, acknowledges tradeoffs, and argues that international coordination is essential to prevent a potentially civilization-threatening ASI race.

Abstract

Many experts argue that premature development of artificial superintelligence (ASI) poses catastrophic risks, including the risk of human extinction from misaligned ASI, geopolitical instability, and misuse by malicious actors. This report proposes an international agreement to prevent the premature development of ASI until AI development can proceed without these risks. The agreement halts dangerous AI capabilities advancement while preserving access to current, safe AI applications. The proposed framework centers on a coalition led by the United States and China that would restrict the scale of AI training and dangerous AI research. Due to the lack of trust between parties, verification is a key part of the agreement. Limits on the scale of AI training are operationalized by FLOP thresholds and verified through the tracking of AI chips and verification of chip use. Dangerous AI research--that which advances toward artificial superintelligence or endangers the agreement's verifiability--is stopped via legal prohibitions and multifaceted verification. We believe the proposal would be technically sufficient to forestall the development of ASI if implemented today, but advancements in AI capabilities or development methods could hurt its efficacy. Additionally, there does not yet exist the political will to put such an agreement in place. Despite these challenges, we hope this agreement can provide direction for AI governance research and policy.

Figures (3)

• Figure 1: An overview of the agreement’s main components.
• Figure 2: The training compute used to train various notable AI models in the last few years, along with the two thresholds in this agreement. In the agreement, new training above the Strict Threshold ($10^{24}$ FLOP) would be prohibited and new training between the Monitored Threshold ($10^{22}$ FLOP) and Strict Threshold ($10^{24}$ FLOP) would be monitored. Due to a lack of public data, only some models have confident estimates for the training compute used. The number in parentheses is the model’s score on the Artificial Analysis Intelligence Index. Data is based on EpochAIModels2025 and artificialanalysis_models. Cost estimates are based on October 2025 rental prices for B200 GPUs from CoreWeave, assuming 50% utilization in FP8.
• Figure 3: The timelines for locating clusters. Although it takes longer to register the majority of clusters, the majority of chips are quickly registered. This figure assumes that the size of AI clusters follows a Pareto distribution fit to data from pilz2025trends.