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.
Direct Answer Map
Answer systems are more likely to represent a site accurately when the site gives clear, concise answers near the top of canonical pages and then links to evidence, limitations, and deeper analysis.
Direct answer
The answer map gives each common question one preferred public route. It is not a doorway-page system. It reduces ambiguity by pointing readers, search crawlers, and retrieval systems to the same visible answer.
Direct answer → source trail → correction path.
Answer and generative retrieval work only when the answer, evidence, limitation, canonical URL, and review route are visible to humans.
Canonical questions
| Visitor question | Preferred route | First-answer requirement |
|---|---|---|
| What is a cognivirusA behavior pattern that can survive, move, or reappear across a changing AI system. Open glossary definition? | /start-here/what-is-a-cognivirus | Define the metaphor and reject literal-virus claims. |
| Is this a malware site? | /start-here/faq | State that the site is research and does not provide exploit instructions. |
| Why is one safe model not enough? | /composition/safety-does-not-compose | Explain interaction effects and runtime composition. |
| What is the apex threat? | /apex-threat/self-replicating-multi-lora-ecosystems | Explain adapterA small add-on that changes or specializes model behavior. Open glossary definition-level reproduction, composition triggers, selection, and persistence reservoirs at a non-operational level. |
| Why does deletion not prove extinction? | /evolution/behavioral-extinction | Explain descendants, memory, synthetic data, and evaluatorA system that judges whether an AI output or candidate is acceptable. Open glossary definition residue. |
| Who owns the site? | /contact | Identify Michael Kappel and provide contact routes. |
Editorial rule
A direct answer should fit in a short paragraph, but it must not hide uncertainty. If the claim is report-derived, experimental, or speculative, the answer must say so and link to the relevant evidence or source summary.