Animated GraphsStrong architectural inferencev1.21.5

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.

Recursive Mitosis-like Branching

Evidence levelStrong architectural inferenceTechnical label: Architectural inference

This graph shows the core A behavior pattern that can survive, move, or reappear across a changing AI system. Open glossary definition concern: behavior can persist while carriers change. It is a conceptual defensive schematic, not a build guide.

animated schematic · behavior persists, carriers change

Recursive mitosis-like branching

This graph shows why the transition graph matters. A behavior can be selected, pruned, inherited, routed, and re-expressed across descendant lineages, while governance must retain a kill switch and lineage evidence.

Recursive mitosis-like branching schematic: a parent model stack branches across generations under selection pressure while inherited adapters, copied prompts, memory inheritance, synthetic-data residue, router exposure, and evaluator scoring create a transition graph.
Static image shown while the animation loads.
The visual emphasizes the Cognivirus thesis: behavior can persist across generations even when carriers, weights, routes, and implementations change.
\n
detail page context

Detail for recursive mitosis-like branching

This page uses the full-width media layout to show how behavior can remain expressible across generations even when carriers, routes, scores, and implementations change.

What the full page adds

  • Selection pressure can preserve patterns that score well even when their origin becomes hard to see.
  • Pruning one branch does not prove related behavior has disappeared from memory, data, prompts, or routes.
  • Containment requires lineage revocation, rollback evidence, monitoring, and behavioral-extinction review.

Review surfaces in the image

generation chaininherited adapterscopied promptsmemory inheritancesynthetic-data residueroute exposureevaluator score

What to notice

Back to all animated graphs · Read how the threat works