Technical ResearchReasoned from system designv1.15.0
In plain English
This page preserves research summaries and source notes. Summaries distinguish direct findings from Cognivirus.com interpretation.
- 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.
Technical Research: Model Ecology, Composition Risk, and AI Governance
Direct answer
This section preserves the deep technical material. It groups the older expert sections under plain labels so first-time readers are not forced to understand every term before navigating.
Technical research paths
- Most Likely Threat Model: distributed behavioral persistence through adapters, routes, memory, evaluators, data, descendants, and human workflows.
- Anatomy of a Cognivirus: where behavior can live: models, prompts, memory, adapters, datasets, tools, evaluators, routes, and people.
- Composition Risk: why safe parts can produce unsafe wholes when combined.
- Evolutionary AI Loops: how updates, tests, selection, memory, and release pressure can preserve behavior over time.
- Control-Plane Risk: why governance is necessary, and why the governance layer must also be reviewed.
- Apex Threat Model: self-replicating multi-LoRA ecosystems, adapterA small add-on that changes or specializes model behavior. Open glossary definition reproduction, and transition-graph risk.
- Evidence System: claim cards, evidence labels, source limits, and what is still unknown.
- Research Notes: source summaries, report syntheses, and preserved research metadata.
- Reference: schemas, catalogs, templates, metrics, and machine-readable records.
- Risk Lab: browser-side worksheets for composition, assurance decayProposed Cognivirus terminology for the loss of confidence in an evaluation result as system components, routes, permissions, models, prompts, memory, tools, or environments change. Open glossary definition, rollback, and responsibility mapping.
What changed in this release
The technical pages remain available. The top navigation now leads with plain-language routes, while this page provides the bridge into expert material.
Suggested path for engineers
- Start with The Problem.
- Read Most Likely Threat Model.
- Review Composition Risk.
- Inspect Evolutionary AI Loops.
- Read The Control-Plane Paradox.
- Use the Risk Lab worksheets.
v1.15 danger-model expansion
- The Cognivirus Danger Model — how distributed persistence, composition, selection, feedback, action, observability, retirement, and rollbackReturning a system to an earlier known state. Open glossary definition fit together.
- Danger Lifecycle — seed, local passA part looks safe by itself. Open glossary definition, composition, expression, selection, residue, inheritance, amplification, retirement gap, and reappearance.
- Action-Layer Risk — why tool access is the hard boundary between weird output and material harm.
- Observability and Replay — why replayable traces are evidence.
- Warning Signals — what operators should watch for.