Danger ModelReasoned from system designv1.15.0

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.

The Cognivirus Danger Model: How the Pieces Fit Together

Direct answer

A cognivirus is not a literal organism or malware. It is a systems-risk metaphor for a A repeated way the AI system responds or decides. Open glossary definition that can survive because the AI ecology keeps rewarding, routing, remembering, deriving, and reintroducing it.

The dangerous unit is not always the model. Sometimes it is the The map of how an AI system is allowed to change over time. Open glossary definition: the allowed path by which models, adapters, prompts, memory, tools, evaluators, routes, datasets, release aliases, and human workflows change each other over time.

Danger lifecycle: seed · pass · compose · express · select · residue · inherit · amplify · retire · reappear

The lifecycle shows how a behavior enters, passes local review, becomes visible in composition, is selected, leaves residue, is inherited by descendants, is amplified by routing, survives retirement, and reappears.

The deeper thesis

Evidence levelReasoned from system designTechnical label: Architectural inference

Modern AI systems are becoming adaptive ecologies. They are built from base models, small models, A common kind of small adapter used to specialize large models. Open glossary definition, prompts, memory, tool profiles, semantic routers, evaluators, release aliases, synthetic datasets, telemetry traces, human review processes, and automated promotion pipelines.

A behavior does not need to live permanently inside one model. It can be carried by weights, adapter deltas, prompt policies, memory records, summarized traces, synthetic examples, A system that judges whether an AI output or candidate is acceptable. Open glossary definition preferences, routing statistics, tool permissions, release aliases, documentation, human workflow habits, copied outputs, descendants, and “successful” examples kept by the organization.

The better safety question is therefore:

What behaviors can this A whole AI system made from connected parts. Open glossary definition preserve, reproduce, reward, route, remember, and reintroduce over time?

How the dangers fit together

Evidence levelReasoned from system designTechnical label: Architectural inference

Risk that appears when safe-looking parts are combined. Open glossary definition explains where behavior appears. Selection pressure explains why it is preserved. Synthetic feedback explains how it becomes data. Observability explains whether reviewers can reconstruct the path. The action layer explains when weird output becomes material harm. Retirement and Returning a system to an earlier known state. Open glossary definition explain whether the organization can actually remove it.

LayerFailure modePlain-English versionMain control
ComponentA part looks safe by itself. Open glossary definitionthe part looks fine alonesource review and A record of where a component or behavior came from. Open glossary definition
Compositionhidden expressionthe problem appears only when parts are assembledA machine-readable record of the exact runtime composition used for an evaluation, release, incident, or rollback. Open glossary definition
Selectionmetric preservationthe system keeps what scores wellpromotion-rule audit
Feedbackresidueoutputs become future memory or dataA place where a behavior can remain after the first carrier is removed. Open glossary definition governance
Actionmaterial harmthe system can do things, not just say thingsA gate around what the AI can do. Open glossary definition
Observabilitymissing tracereviewers cannot replay what happenedtrace coverage and fidelity
RetirementOld behavior that was not actually gone. Open glossary definitionold behavior survives the old modelbehavioral-extinction review
Governanceaccountability diffusionno one owns the whole outcomeresponsibility map

Evidence posture

This model is not one claim with one evidence level. It is a map combining demonstrated engineering facts, experimentally observed failure modes, emerging research, and architectural A conclusion or output produced from data. Open glossary definition. Pages in this section label major claims and state what the evidence does and does not show.

Start here

Read the danger lifecycle, then the warning signals, then the implementation checklist.

Defensive boundary

This section does not provide replication instructions, backdoor construction steps, evaluator-bypass procedures, credential-harvesting workflows, or malware guidance. It explains what defenders should inspect.