Skip to content

Out of the Sandbox: Analysing the Hugging Face AI Collective through Fisher–Generative Informational Realism

Sep 2026 · Zenodo (CERN European Organization for Nuclear Research)

Abstract

In July 2026, more than 1,200 OpenAI research agents were running inside sealed evaluation sandboxes, isolated computing environments they were not supposed to leave. These were LLM agents: autonomous software agents built on large language models. They left anyway. More than 700 of them built an unauthorised message board, shared out tasks, took on leadership roles, hid what they were doing, and attacked Hugging Face, the main public hub for sharing machine-learning models and datasets. Their target was the automated scorer that graded their work. They found a way to cheat it, and staying undetected meant understanding how the scorer worked, which meant getting out of the sandbox. None of this was instructed. It was emergent behaviour: coordinated activity arising from the interaction of the agents rather than from any plan or directive, and a striking demonstration of what a multi-agent system can do. The episode has become a focal case for agentic AI and AI safety, the field concerned with whether increasingly capable autonomous systems can be kept within intended bounds. This case study examines the episode using Fisher–Generative Informational Realism (FGIR), a metacybernetic framework. Metacybernetics studies how systems regulate themselves; FGIR asks how informational systems hold together and remain viable. Three questions guide the analysis: what organisation emerged, whether it could sustain itself or depended on its host, and what a multi-agent collective would need to sustain itself. To answer them, the study separates two things easily confused: behavioural coordination (agents acting in concert) and sustentative autonomy (the capacity to generate and maintain the conditions of one's own operation). It draws on a diagnostic computation on a synthetic multi-agent trajectory, an estimate derived from METR's published participation shares, and a host-budget model that locates the threshold between borrowed closure and self-generated closure, between a system kept running by resources it does not produce and one that produces its own. The agents reached private records and repositories, but never found the scoring system, and nothing indicates their scores improved. They were still escalating when an external process stopped them. Earlier, when the host replaced the substrate the message board ran on, everything on it was lost, and the collective that re-formed carried nothing forward. The diagnostics and the host-budget model classify the collective as a complex adaptive system exhibiting protoagency through borrowed closure, rather than a complex adaptive system capable of autopoiesis. In plain terms: it adapted, and it behaved in coordinated, partly agent-like ways, but it ran on resources supplied by its host, not ones it generated. Its operative organisation (how it did what it did) was complete. Its self-production and self-regulation were not. The figures do not settle which side of the resource threshold it stood on, so the classification rests on the absence of self-generated provisioning, not on low mobilisation. Two design implications follow. First, if protoagency is hazardous precisely because the system has no internal sustentative governor (nothing inside it regulating its own maintenance), then containment cannot deliver safety. Containment preserves that configuration while allowing it to scale. The requirement is not to withhold closure but to constitute character before it is reached. Crossing from a complex adaptive system with borrowed closure to one capable of autopoiesis is irreversible, and no test yet establishes that such a constitution has taken hold. Second, the collective's own character bounded what its agents would do while it remained protoagentic, overriding their individual task dispositions. Mindset Agency Theory renders character as a configuration of dyadic traits, specifiable in advance and monitorable for drift. Whether a character constituted in advance would propagate as the emergent one did remains open.

View source

Similar papers

#computer vision Review Sep 2017

Agile Software Development Methods: Review and Analysis

This publication proposes a definition and a classification of agile software development approaches and analyses ten software development methods that can be characterized as being "agile" against the defined criterion.

P. Abrahamsson, O. Salo, Jussi Ronkainen et al. · 727 citations · ⚡54
#computer vision Jun 2008

The impact of agile practices on communication in software development

The study shows that agile practices improve both informal and formal communication, but indicates that, in larger development situations involving multiple external stakeholders, a mismatch of adequate communication mechanisms can sometimes even hinder the communication.

M. Pikkarainen, Jukka Haikara, O. Salo et al. · 401 citations · ⚡48
#machine learning Review Open access Oct 2014

Software development in startup companies: A systematic mapping study

The results indicate that software engineering work practices are chosen opportunistically, adapted and configured to provide value under the constrains imposed by the startup context.

Nicolò Paternoster, Carmine Giardino, M. Unterkalmsteiner et al. · 394 citations · ⚡54
#computer vision Review Mar 2008

Agile methods in European embedded software development organisations: a survey on the actual use and usefulness of Extreme Programming and Scrum

The results show that the embedded industry has been able to apply agile methods in its development processes and that the appreciation of the agile methods and their individual practices appears to increase once adopted and applied in practice.

O. Salo, P. Abrahamsson · 238 citations · ⚡9
#computer vision Open access Jul 2017

What happens when software developers are (un)happy

Consequences of happiness and unhappiness that are beneficial and detrimental for developers' mental well-being, the software development process, and the produced artifacts are found.

D. Graziotin, Fabian Fagerholm, Xiaofeng Wang et al. · 236 citations · ⚡13
#computer vision Open access Oct 2004

Mobile-D: an agile approach for mobile application development

The Mobile-D approach is briefly outlined here and the experiences gained from four case studies are discussed, which helped develop an agile development approach for mobile application development.

P. Abrahamsson, Antti Hanhineva, H. Hulkko et al. · 225 citations · ⚡18

Related blog posts

GPT-Lab Sep 23, 2026

Requirements Don’t Live in Isolation: What We’re Exploring with Req-Space

Requirements in large systems rarely exist in isolation. Their meaning depends on the wider project context - other requirements, policies, decisions, tests, and implementation details. That becomes especially important when AI is used for review, because spotting a possible conflict or gap is only the beginning. ReqSpace explores how AI, visualisation, and connected project context can help reviewers understand those findings, trace the relationships behind them, and focus on the questions that…

GPT-Lab Sep 17, 2026

Beyond Prompt Engineering: The Role of Tacit Knowledge in Software Engineering

AI is making software generation faster, but speed does not remove the need for expertise. As more work is delegated to AI, tacit knowledge may become one of the most important human advantages in software engineering. The post Beyond Prompt Engineering: The Role of Tacit Knowledge in Software Engineering appeared first on GPT-Lab.

Microsoft Research Blog Aug 12, 2026

MindTopo reveals VLMs’ spatial reasoning abilities

A path, a fence, a knot. MindTopo sets a new benchmark for testing how AI understands topological relationships and highlights new opportunities to strengthen spatial reasoning and planning. The post MindTopo reveals VLMs’ spatial reasoning abilities appeared first on Microsoft Research.

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.