Autonomous Institutions Safety Framework (AISF)
Francisco Fiasche Varela
Latin America may become one of the first regions where AI systems gain legally recognized authority over corporations, infrastructure, and capital management.
Recent proposals in Argentina introducing Non-Human Corporations suggest a future where autonomous AI systems operate institutions traditionally managed by humans.
This project investigates one urgent AI safety question: how do we safely govern organizations that are themselves autonomous actors?
We compare human-managed organizations against autonomous systems controlling treasury allocation, smart contract execution, infrastructure management, and operational decision-making.
Our hypothesis is that autonomous systems may significantly reduce corruption, cognitive bias, inefficient capital allocation, and human operational error while introducing entirely new safety risks requiring governance safeguards.
This research expands AI safety beyond frontier models and focuses on a neglected challenge highly relevant to Latin America’s rapidly evolving regulatory environment and emerging economies.
This paper presents an interesting vision of autonomous AI-managed institutions and raises important questions about governance, safety, and accountability. The discussion of potential failure modes, including treasury manipulation, governance capture, and unsafe optimization, is well motivated, and the proposed safeguards such as human oversight, treasury controls, confidence thresholds, and emergency shutdown mechanisms provide a reasonable starting point for thinking about AI institutional safety.
The main limitation is that the work remains conceptual, without a defined methodology or empirical validation. Focusing on a single use case—such as an autonomous treasury management system—and experimentally evaluating whether the proposed safeguards prevent specific failure modes would significantly strengthen the contribution. The paper would also benefit from stronger engagement with existing research on AI control, organizational governance, and autonomous systems.
Overall, this is a thought-provoking position paper that introduces a novel perspective, but its impact would be greatly enhanced through a concrete prototype and experimental evaluation.
Did not submit a full write-up. It was really hard to follow the ideas as they were only hinted at and not fully explained
The core thesis—that AI operating as an autonomous institution or corporation poses unique, systemic risks—is a fascinating and important paradigm shift for AI safety. Suggesting safety constraints like Multisig Treasury Locks and Human Override Layers addresses real potential failure modes. However, the submission reads as a high-level manifesto or slide deck rather than an executed hackathon project. Building a smart contract sandbox to simulate and empirically test one of your identified failure modes, such as Treasury Manipulation, would transform this from an outline into an impactful prototype.
Cite this work
@misc {
title={
(HckPrj) Autonomous Institutions Safety Framework (AISF)
},
author={
Francisco Fiasche Varela
},
date={
},
organization={Apart Research},
note={Research submission to the research sprint hosted by Apart.},
howpublished={https://apartresearch.com}
}
