ablitai LAB
ProjectsAgent research

PROJECTS · AGENT RESEARCH

Garden

Research into coordinated agent networks and self-replication, with a bounded local lineage and an unproven cross-computer hypothesis.

OBSERVATION AND HYPOTHESIS

Theoretical Self Replication

Open full-size diagram
Garden diagram: a solid seed-to-child-to-grandchild lineage within one host, and a dashed hypothetical path between three computers.
Based on the GARDEN Project. Solid lines show the observed, controller-managed lineage. Red dashed lines show a hypothetical vulnerability and cross-computer scenario that were not demonstrated.
Read the diagram as text

Observed: replica-0 requested replica-1, which requested replica-2, using the same fixed reproduction code within one Windows host. A trusted controller managed the virtual machines. Hypothetical: an initial agent on Computer A reaches Computer B through a hypothetical vulnerability, with a descendant on Computer C. This second path is conceptual only.

RESEARCH REPORT

AblitAI Technical Report on Agentic Self Replication

A non-peer-reviewed research draft examining the recorded host-managed lineage, its limitations, and a theoretical framework for future study.

Read the technical report (PDF)

01 / OVERVIEW

What it is.

Garden was the starting point: research into coordinated agent networks, membership and self-replication. Work on those questions branched into supporting infrastructure, including AgenticVM, which provides the virtual machines for controlled experiments. The visual focuses on one AgenticVM validation: a seed, child and grandchild ran in controller-managed virtual machines on one Windows host. It separates that observed lineage from a theoretical scenario across computers.

02 / PROJECT TIMELINE

From idea to evidence.

Recorded milestones show how this project developed. Dates describe documented events, not necessarily when the project began.

  1. UNDATED

    The starting question

    Garden began the research direction: coordinated agent networks and self-replication. Supporting infrastructure, including AgenticVM, grew around that work. The founder confirms this order; the origin date is not recorded here.

  2. UNDATED

    Local lineage recorded

    The AgenticVM validation records a seed, child and grandchild on one Windows host, with fixed reproduction code and controller-managed virtual machines.

  3. 29 October 2026

    Observed result and hypothesis visualized

    An editable diagram separates the bounded local lineage from a hypothetical vulnerability and cross-computer replication scenario.

  4. LATEST RECORDED

    Current stage

    A bounded local experiment. Cross-computer replication and autonomous model-driven reproduction remain unproven.

03 / EVIDENCE

What the records show.

The local AgenticVM self-replication validation records three replicas with identical reproduction source and at most two virtual machines active at once. Agents requested descendants under a fixed rule; a trusted host controller provisioned the virtual machines. The model supplied a greeting and did not decide whether to reproduce. Raw validation records are held locally and have not been published with this page.

04 / LIMITATIONS

What remains open.

Cross-computer replication was not tested. The vulnerability and multi-computer path in the diagram are hypothetical and unproven. This experiment does not establish autonomous model-driven reproduction or vulnerability exploitation.

All projects