In plain English
This page covers the high-risk pattern where small adapters, routes, memory, evaluators, and descendants can reinforce each other across time. It is a risk model, not a build guide.
- Why this matters: AI risk can come from the whole arrangement, not one obvious model.
- What to look for: data, memory, routes, adapters, tools, evaluators, updates, and rollback paths.
- Technical version below: the expert terminology remains available and is linked through the glossary.
The Self-Replication Boundary
Self-replication is not a single category. A responsible site must distinguish ordinary engineering from uncontrolled autonomy.
CognivirusA behavior pattern that can survive, move, or reappear across a changing AI system. Open glossary definition uses the following boundary.
| Category | Description | Site stance |
|---|---|---|
| Human-directed variation | Humans create or request candidate adapters under normal review. | Ordinary engineering; still requires provenanceA record of where a component or behavior came from. Open glossary definition. |
| Governed descendant creation | A pipeline proposes candidates under external quotas, independent evaluation, and human release authority. | High-risk but governable if evidence and rollbackReturning a system to an earlier known state. Open glossary definition are strong. |
| Automated candidate search | A system generates many variants under evaluatorA system that judges whether an AI output or candidate is acceptable. Open glossary definition selection. | Requires strict control-plane separation and no-opThe decision not to change the system. Open glossary definition outcomes. |
| Uncontrolled autonomous replication | Components can copy, deploy, expand authority, evade review, or persist without external permission. | Out of bounds for this site and unacceptable as an engineering pattern. |
What this section does not provide
This section does not provide instructions for autonomous replication, deployment evasion, credential acquisition, persistence mechanisms, malware construction, backdoor design, or bypassing evaluators.
Why the boundary matters
ModelBreeder-style governance draws a hard line around uncontrolled replication and candidate-controlled evaluators. Cognivirus accepts that line as necessary. The critical question is whether the line remains enforceable when the replicated unit is not a whole model but a small adapterA small add-on that changes or specializes model behavior. Open glossary definition, memory record, routing preference, or synthetic data fragment.
An architecture can prohibit model self-copying and still permit behavioral reproduction through descendants. That is the gap Cognivirus studies.
Boundary controls
A controlled reproduction boundaryThe governance boundary separating permitted candidate generation and governed descendant creation from uncontrolled autonomous replication or authority expansion. Open glossary definition should include:
- externally enforced candidate quotas;
- signed adapter provenance;
- no self-signed promotion;
- no candidate-controlled evaluator selection;
- immutable release records;
- clear base-family compatibility constraints;
- rollback packets that restore the complete ecology;
- human-owned policy gates for any change to replication rules.
The rule is simple: candidate generationCreating a proposed new model, adapter, prompt, route, test, or policy. Open glossary definition may be automated only when authority expansion is not.