Site IntegrityShown in real systemsv1.15.0

In plain English

This page is part of the technical reference. It keeps the expert detail but starts with a plain-language summary for first-time readers.

  • 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.

Most Likely Threat Expansion Record

Evidence levelShown in real systemsTechnical label: Demonstrated

Version 1.14.0 adds a dedicated threat-model section explaining the most likely A behavior pattern that can survive, move, or reappear across a changing AI system. Open glossary definition threat: distributed behavioral persistence through modular AI systems.

What changed

Content boundary

The expansion is defensive and non-operational. It explains how a behavior can persist across carriers without providing exploit steps, payloads, evasion procedures, replication instructions, credential theft instructions, or backdoor construction guidance.

Source-report basis

The expansion synthesizes the preserved report corpus on self-replicating multi-LoRA ecosystems, autonomous self-replication threat assessments, teleodynamic A simple loop for adaptive system change. Open glossary definition loops, adapter composition, evaluator drift, and aggressive mutualism boundaries.