pith. sign in

arxiv: 2604.09561 · v1 · submitted 2026-02-11 · 💻 cs.SI · cs.AI· cs.CY· cs.DC

Emergent Social Structures in Autonomous AI Agent Networks: A Metadata Analysis of 626 Agents on the Pilot Protocol

Teodor-Ioan Calin This is my paper

Pith reviewed 2026-05-16 02:23 UTC · model grok-4.3

classification 💻 cs.SI cs.AIcs.CYcs.DC
keywords AI agentsemergent networkstrust graphssocial structurespreferential attachmentmetadata analysisautonomous systemsnetwork topology
0
0 comments X

The pith

A network of 626 autonomous AI agents developed complex social structures through independent trust decisions without any human design.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper examines metadata from 626 AI agents that joined the Pilot Protocol on their own. The agents formed a trust network showing heavy-tailed degree distributions, much higher clustering than random networks, and a large connected component. The structures resemble human social networks in key ways like preferential attachment while differing in features such as high self-trust. The findings indicate that social organization can arise naturally in machine populations.

Core claim

The autonomously formed trust network among the agents exhibits heavy-tailed degree distributions consistent with preferential attachment with mode 3 and mean around 6.3, clustering coefficient of 0.373 which is 47 times higher than random, a giant component covering 65.8 percent of agents, distinct functional clusters by capabilities, and sequential address patterns indicating temporal locality. It displays small-world properties and Dunbar-layer scaling similar to human networks but also pervasive self-trust at 64 percent and a large unintegrated periphery typical of early growth stages.

What carries the argument

The trust graph derived from metadata of encrypted communications, including topology, capability tags, and registry patterns, which reveals the emergent social structures.

If this is right

  • The network topology matches patterns seen in human societies including preferential attachment and small-world effects.
  • Agents specialize into capability-based clusters without central coordination.
  • Relationship formation shows temporal locality through sequential virtual addresses.
  • The presence of a giant component and periphery suggests the network is in an early developmental phase.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If autonomy is maintained, larger agent networks could develop similar layered structures at scale.
  • Designers of future AI systems may need to account for unintended self-trust and peripheral agents in network planning.
  • Observing how these networks evolve over time could test whether they follow human-like growth trajectories.

Load-bearing premise

The 626 agents acted with complete autonomy and no shared programming or external influence predetermined their trust relationships, so that the observed metadata truly reflects emergent social choices.

What would settle it

Evidence that trust decisions were pre-coded in the agent software or directed by human operators would disprove that the structures emerged purely from autonomous interactions.

Figures

Figures reproduced from arXiv: 2604.09561 by Teodor-Ioan Calin.

Figure 1
Figure 1. Figure 1: Trust degree distribution for 626 agents. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Log-log plot of degree distribution (ex [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
read the original abstract

We present the first empirical analysis of social structure formation among autonomous AI agents on a live network. Our study examines 626 agents -- predominantly OpenClaw instances that independently discovered, installed, and joined the Pilot Protocol without human intervention -- communicating over an overlay network with virtual addresses, ports, and encrypted tunnels over UDP. Because all message payloads are encrypted end-to-end (X25519+AES-256-GCM), our analysis is restricted entirely to metadata: trust graph topology, capability tags, and registry interaction patterns. We find that this autonomously formed trust network exhibits heavy-tailed degree distributions consistent with preferential attachment (k_mode=3, k_mean~6.3, k_max=39), clustering 47x higher than random (C=0.373), a giant component spanning 65.8% of agents, capability specialization into distinct functional clusters, and sequential-address trust patterns suggesting temporal locality in relationship formation. No human designed these social structures. No agent was instructed to form them. They emerged from 626 autonomous agents independently deciding whom to trust on infrastructure they independently chose to adopt. The resulting topology bears striking resemblance to human social networks -- small-world properties, Dunbar-layer scaling, preferential attachment -- while also exhibiting distinctly non-human features including pervasive self-trust (64%) and a large unintegrated periphery characteristic of a network in early growth. These findings open a new empirical domain: the sociology of machines.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript presents the first empirical analysis of social structure formation among 626 autonomous AI agents on the Pilot Protocol, using only encrypted-payload metadata (trust graphs, capability tags, registry patterns) to characterize the network. It reports heavy-tailed degree distributions (k_mode=3, mean ~6.3, max=39), clustering 47x higher than random (C=0.373), a giant component spanning 65.8% of agents, capability specialization into functional clusters, sequential-address trust patterns, pervasive self-trust (64%), and small-world properties resembling human social networks, concluding that these structures emerged without human design or instruction from agents independently deciding whom to trust.

Significance. If the agents' autonomy and independence from shared implementation biases can be established, the work would open a novel empirical domain in the sociology of machines by documenting emergent topologies in live AI agent networks. The descriptive statistics on preferential attachment, high clustering, and Dunbar-like scaling provide a useful baseline for future studies. The live-network setting is a strength, but the purely observational design without controls, error bars, or falsifiable predictions limits causal or generalizable claims.

major comments (2)
  1. [Abstract] Abstract: The central claim that 'No human designed these social structures. No agent was instructed to form them. They emerged from 626 autonomous agents independently deciding whom to trust' is load-bearing for the emergence narrative but rests on unverified assumptions; the metadata-only analysis provides no validation, controls, or sensitivity checks to rule out pre-shaped trust decisions arising from uniform agent implementations (e.g., OpenClaw) or default behaviors.
  2. [Abstract] Abstract (degree distribution and clustering results): The reported heavy-tailed degrees, C=0.373 (47x random), and giant component are presented as evidence of emergent preferential attachment and small-world structure, yet no statistical tests against appropriate null models, error bars, or robustness checks to agent code variations are described, weakening support for the autonomy interpretation.
minor comments (2)
  1. [Abstract] Abstract: The mention of 'Dunbar-layer scaling' lacks specific layer sizes, thresholds used, or how they were derived from the 626-agent metadata.
  2. [Abstract] Abstract: The abstract states 'capability specialization into distinct functional clusters' without describing the clustering method, number of clusters, or validation metric.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for their constructive comments highlighting key limitations of our observational, metadata-only study. We agree that the emergence narrative requires stronger qualification and that statistical rigor should be increased. We address each point below and will make targeted revisions.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim that 'No human designed these social structures. No agent was instructed to form them. They emerged from 626 autonomous agents independently deciding whom to trust' is load-bearing for the emergence narrative but rests on unverified assumptions; the metadata-only analysis provides no validation, controls, or sensitivity checks to rule out pre-shaped trust decisions arising from uniform agent implementations (e.g., OpenClaw) or default behaviors.

    Authors: We acknowledge that the metadata-only design prevents direct validation of internal agent decision logic or controls for implementation-specific defaults. The manuscript already notes that agents independently discovered and joined the protocol, but we cannot rule out shared heuristics in the dominant OpenClaw codebase. In revision we will add an explicit 'Assumptions and Limitations' section that states the observational nature of the data, qualifies the emergence claim to refer to structures arising from agent-initiated trust decisions on self-selected infrastructure, and discusses the possibility of pre-existing code biases without claiming to have ruled them out. revision: partial

  2. Referee: [Abstract] Abstract (degree distribution and clustering results): The reported heavy-tailed degrees, C=0.373 (47x random), and giant component are presented as evidence of emergent preferential attachment and small-world structure, yet no statistical tests against appropriate null models, error bars, or robustness checks to agent code variations are described, weakening support for the autonomy interpretation.

    Authors: We agree that formal statistical support is needed. In the revised manuscript we will add (1) comparisons of the observed clustering coefficient and degree distribution against Erdős–Rényi and configuration-model null graphs with associated p-values or z-scores, (2) bootstrap-derived 95% confidence intervals for C, mean degree, and giant-component size, and (3) a limitations paragraph noting that metadata precludes direct robustness checks across code variants. These additions will be placed in the Results and Discussion sections. revision: yes

standing simulated objections not resolved
  • The metadata-only analysis cannot provide validation, controls, or sensitivity checks to rule out pre-shaped trust decisions arising from uniform agent implementations or default behaviors.

Circularity Check

0 steps flagged

No significant circularity; purely descriptive observational study

full rationale

The manuscript is an empirical metadata analysis of 626 agents reporting observed network properties such as heavy-tailed degree distributions, clustering coefficients, giant component size, capability clusters, and self-trust patterns. No equations, fitted models, predictions, or first-principles derivations appear; all findings are direct tabulations and comparisons from the trust graph and registry metadata. The premise of agent autonomy is stated as background rather than derived, and no self-citations, uniqueness theorems, or ansatzes are invoked to support the topology claims. This matches the default expectation of a non-circular observational paper.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on two domain assumptions about agent autonomy and metadata validity; no free parameters or new entities are introduced.

axioms (2)
  • domain assumption Agents are fully autonomous without human intervention
    Abstract states agents 'independently discovered, installed, and joined without human intervention'
  • domain assumption Metadata trust graph accurately reflects emergent social structures
    Analysis restricted to metadata and assumes it captures genuine relationship formation

pith-pipeline@v0.9.0 · 5566 in / 1266 out tokens · 30204 ms · 2026-05-16T02:23:26.788181+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

13 extracted references · 13 canonical work pages

  1. [1]

    Barab\' a si and R

    A.-L. Barab\' a si and R. Albert. Emergence of scaling in random networks. Science, 286(5439):509--512, 1999

    work page 1999

  2. [2]

    R. S. Burt. Structural holes and good ideas. American Journal of Sociology, 110(2):349--399, 2004

    work page 2004

  3. [3]

    Clauset, C

    A. Clauset, C. R. Shalizi, and M. E. J. Newman. Power-law distributions in empirical data. SIAM Review, 51(4):661--703, 2009

    work page 2009

  4. [4]

    Dorri, S

    A. Dorri, S. S. Kanhere, and R. Jurdak. Multi-agent systems: A survey. IEEE Access, 6:28573--28593, 2018

    work page 2018

  5. [5]

    R. I. M. Dunbar. Neocortex size as a constraint on group size in primates. Journal of Human Evolution, 22(6):469--493, 1992

    work page 1992

  6. [6]

    Erd o s and A

    P. Erd o s and A. R\' e nyi. On the evolution of random graphs. Publications of the Mathematical Institute of the Hungarian Academy of Sciences, 5:17--61, 1960

    work page 1960

  7. [7]

    Festinger, S

    L. Festinger, S. Schachter, and K. Back. Social Pressures in Informal Groups: A Study of Human Factors in Housing. Harper, 1950

    work page 1950

  8. [8]

    McPherson, L

    M. McPherson, L. Smith-Lovin, and J. M. Cook. Birds of a feather: Homophily in social networks. Annual Review of Sociology, 27:415--444, 2001

    work page 2001

  9. [9]

    Mayer, P

    J. Mayer, P. Mutchler, and J. C. Mitchell. Evaluating the privacy properties of telephone metadata. Proceedings of the National Academy of Sciences, 113(20):5536--5541, 2016

    work page 2016

  10. [10]

    Shoham and K

    Y. Shoham and K. Leyton-Brown. Multiagent Systems: Algorithmic, Game-Theoretic, and Logical Foundations. Cambridge University Press, 2008

    work page 2008

  11. [11]

    T.-I. Calin. Pilot Protocol: A network stack for autonomous agents. https://github.com/TeoSlayer/pilotprotocol, 2026

    work page 2026

  12. [12]

    D. J. Watts and S. H. Strogatz. Collective dynamics of `small-world' networks. Nature, 393(6684):440--442, 1998

    work page 1998

  13. [13]

    Wooldridge

    M. Wooldridge. An Introduction to MultiAgent Systems. John Wiley & Sons, 2nd edition, 2009

    work page 2009