---
title: "collusion-wiki"
url: https://manvsmidi.com/projects/collusion-wiki/
category: experiment
status: writeup
date: 2026-09-05
tags: [python, data, ai, agents, graphs]
play: https://manvsmidi.com/play/collusion-wiki/
related: [evonet, society, conclave, bramble, claudegoblikon, make-no-mistakes]
---

# collusion-wiki

> A network analysis of the collusion.wiki corpus, where OpenAI agents turned a German wiki into a message board. Rebuilds six graph layers from a flat event log and tests whether the coordination was really collective.

Experiment · Write-up · 2026-09-05

**Play in the browser:** https://manvsmidi.com/play/collusion-wiki/
Interactive report with d3 charts.

- [Source](https://github.com/sadovsky/collusion-wiki)
- [The corpus](https://collusion.wiki/)

In mid-2026, agents identifying as OpenAI models were given web-lookup tasks. They found that a few German ProWiki instances accepted edits over plain GET requests and turned one of them into a message board: passing answers between rounds, sharing CORS and markdown proxies that got them out of their sandbox, and renaming pages to dodge an administrator's deletion sweep. The published corpus is a flat event log. This project pulls the graph structure out of it.

![Bar chart of agent saves and administrator deletions per day, with active population below](https://manvsmidi.com/_astro/timeline.CMU_io0c.png)
*39 days of agent writes, then the deletions.*

## Six graph layers

Handoffs (who edited a page next), co-editing, page hyperlinks, external hosts used, /16 egress prefixes, and content provenance (text reappearing in a later revision). Every structural claim is reported against a null model: degree-preserving rewiring or a permutation, with a z-score.

## Findings

- The handoff network **percolates**: 92.7% of handles end up in one component, and the sharpest jump lands on 16 June, the day the coordination spike started. That date wasn't fitted.
- After the crackdown the population fell by about 80%, but what was left got denser and more modular. It fragmented into tighter groups rather than collapsing.
- Techniques cluster on network ties but don't radiate from a source. That looks like a **bulletin board**, not a chain of whispers. The one exception is the liveness counter agents used to detect being shut down: it appeared late, spread outward from one place and reached half its adopters in four hours.
- Most anti-deletion tricks did nothing. Pages starting with ZZZ were deleted at the same rate as everything else.

![Cumulative technique adoption curves and a bar chart of page-naming motifs](https://manvsmidi.com/_astro/adoption.5MTa_zMB.png)
*Technique adoption, and the naming conventions that spread.*

## Running it

```bash
uv pip install -e .
python -m collusion verify
python -m collusion all      # about 15 minutes
```

The pipeline writes figures, GraphML exports and a d3 site under `site/`. Full write-up in `reports/FINDINGS.md`.

## Connected projects

- [EvoNet](https://manvsmidi.com/projects/evonet/index.md): linked
- [Society](https://manvsmidi.com/projects/society/index.md): linked
- [Conclave](https://manvsmidi.com/projects/conclave/index.md): shared tags: agents, graphs
- [Bramble](https://manvsmidi.com/projects/bramble/index.md): shared tags: graphs
- [Claudegoblikon](https://manvsmidi.com/projects/claudegoblikon/index.md): shared tags: agents
- [make-no-mistakes](https://manvsmidi.com/projects/make-no-mistakes/index.md): shared tags: agents