AI Treaty Momentum Index (ATMI)
Oksana Kotelnikova, Sofiia Lobanova · Team AI Treaty Momentum Index
Submitted to The AI Forecasting Hackathon. Sprint projects are early-stage work by participants, not Apart Research publications.
We present the AI Treaty Momentum Index (ATMI), a lightweight forecasting model that estimates the near-term chance of a new international AI risk treaty or binding regulation. ATMI learns from past treaties in climate, nuclear, biosafety, and finance by encoding each case with mechanism classes—buckets of factors that help or hinder agreement (e.g., Policy window, Capability shock, Verification feasibility, Public salience, Epistemic shift, Geopolitical alignment). Each mechanism is scored from −5 (hurts) to +5 (helps) per case, and we map these mechanism profiles to outcomes (treaty vs. no treaty). We use the fit in two modes: Nowcasting—apply today’s AI governance signals to produce a current chance with 80% intervals; Scenario-casting—stress-testing futures where scenarios fork from AI-2027 milestones to see how the odds shift. Early findings: the largest marginal lifts come from Epistemic shift and Geopolitical alignment, followed by Public salience and Epistemic infrastructure. As of Nov 2025, ATMI estimates a ~0.6% chance of a new treaty, suppressed by misaligned geopolitics, weak verification readiness, and negative epistemic/resource shifts. We propose to follow this MVP and build a public “anti-doomsday” tool: a web dashboard that tracks each mechanism with live indicators, shows the ATMI curve over time, and flags actionable weakest links (the marginal mechanisms whose improvement most raises the chance of agreement).
Reviews
* It isn't clear to me what was taken into account to determine the prior medians; given that this is the foundation of your contribution, I'd like to see more justification and explanation of the approach. For example, a case study for one of these would have been quite valuable to see.
* Another way to connect this further to AI cases would have been to examine specific instances in the recent past where any number of these influences occurred, and what the impact on policy was, if anything. Immediately I can think of the cluster of incidents surrounding deepfake porn, which did catalyze some US legislation.
* I feel that the "nowcast" is where the bulk of value is introduced, at least from a policymaker perspective. Given that, I would've liked to see more focus on the inputs used here, possibly even including uncertainty analysis.
I like the topic and the conclusions that come out of the project: That there are 'high-leverage' mechanisms that need to happen before an AI treaty can happen.
On the whole, I think the forecasting aspect of the project is weaker: there are quite a few assumptions (e.g. prior medians, prior range) that sort of are made by the authors that automatically lead to the conclusions as stated. On the whole this is a useful setting for a thought experiment but I would encourage the authors to focus on more realistically modelling some specific criteria (e.g. just on epistemic shift + geopolitical alignment)
Cite this project
@misc{kotelnikova2025ai,
title = {{AI Treaty Momentum Index (ATMI)}},
author = {Oksana Kotelnikova and Sofiia Lobanova},
year = {2025},
month = nov,
note = {Submitted to The AI Forecasting Hackathon, an Apart Research Sprint},
howpublished = {\url{https://apartresearch.com/sprints/projects/ai-treaty-momentum-index-atmi-lst4}},
url = {https://apartresearch.com/sprints/projects/ai-treaty-momentum-index-atmi-lst4}
}More from The AI Forecasting Hackathon
- View project: System Dynamics Game-Theoretic Model of the AI Development Race
System Dynamics Game-Theoretic Model of the AI Development Race
System Dynamics BCN
A Game theoretic / System Dynamics model of the race dynamics of the US, China, and EU, as a follow up to the Armstrong et al. (2016) paper “Racing to the Precipice”. We find preliminary results where knowledge of …
- View project: ExogenousAI
ExogenousAI
Fibonacci
Current AI capability forecasting methodologies, including EpochAI's Direct Approach and Biological Anchors framework, primarily rely on internal metrics such as training compute and scaling laws while assuming stable …
- View project: AI Incidents Forecasting
AI Incidents Forecasting
KLACE
This research develops a framework for forecasting AI incidents to help predict future risks. We have developed two models that forecasts incidents which include calibrated 90% prediction intervals with backtests. These …