Site IntegrityDemonstrated real incidentv1.22.1

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.

ModelBreeder Risk-Side Refocus Closeout v1.22.0

Evidence levelDemonstrated real incidentTechnical label: Demonstrated

This package changes the use of the uploaded ModelBreeder material on A behavior pattern that can survive, move, or reappear across a changing AI system. Open glossary definition. The material remains preserved as source reports, but the public Cognivirus content now uses it primarily to deepen risk analysis rather than possibility framing.

What changed

Carried-forward rules

Validation target

The validation target for this release is that the new pages render, indexes include them, .uai memory contains the risk-side directive, and the package remains pure PHP, JavaScript, CSS, Markdown, and JSON with no new build tool or third-party JavaScript dependency.