EvidenceStrong architectural inferencev1.22.1
In plain English
This page shows what kind of support exists for each claim: real systems, experiments, early evidence, architectural reasoning, open questions, or speculative scenarios.
- 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 no-self-replication boundary
Evidence card
- Claim
- Model breeding can create descendant artifacts without allowing autonomous installation, remote copying, or hidden persistence.
- Evidence level
- Architectural inference
- Source
- https://modelbreeder.com/safety/no-self-replication-boundary
- Publication date
- 2026-06-26
- Authors or institution
- ModelBreeder.com
- System tested
- Safety distinction between governed candidate creation and uncontrolled propagation.
- Limitations
- Does not address functional persistence of behavior through approved descendants.
- What the evidence does show
- Model breeding can create descendant artifacts without allowing autonomous installation, remote copying, or hidden persistence.
- What the evidence does not show
- That absence of literal self-replication prevents behavior from persisting through ordinary engineering succession.
- Date last reviewed in UTC
- 2026-06-26T00:00:00Z
Site use
This source supports Cognivirus.com pages related to replication boundary, governed descendants, release channels. Its role is bounded by the limitations listed above.