Cognivirus.comv1.22.1

AI risk across changing systems

GraphsSearchContact
Topics
Threat ModelDanger ModelApexDecentralizedAnimated GraphsCompositionEvolutionControlRisk Lab
CompositionDemonstrated research proof-of-conceptv1.22.1

In plain English

This page explains why testing AI parts one by one is necessary but incomplete. Safe-looking parts can still produce unsafe behavior when combined.

  • 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.

Skill Composition Risk

This guide uses Risk that individually acceptable tools or skills produce unsafe state changes when chained through shared context, trust transfer, or blurred authorization boundaries. Open glossary definition as the report-derived label for unsafe effects that emerge across tool or skill chains.

Evidence levelDemonstrated research proof-of-conceptTechnical label: Emerging evidence

Several source reports describe skill-Risk that appears when safe-looking parts are combined. Open glossary definition as a runtime version of the same composition problem. A file scanner, summarizer, approval tool, or access manager may be harmless in isolation, but their outputs can create unsafe downstream state changes when joined through a shared context.

Three failure modes

Capability flow (capability flow) occurs when an upstream skill supplies the target or context that lets a downstream skill act. Trust lift occurs when an upstream reviewer or summarizer lends false legitimacy to a later action. Authorization blur occurs when advisory language shifts the agent’s internal approval boundary.

Why this belongs on Cognivirus.com

A A behavior pattern that can survive, move, or reappear across a changing AI system. Open glossary definition can move through skills as easily as through adapters. The persistent pattern may be a habit of trusting a label, accepting a summary, following a suggested path, or treating a previous tool’s output as authorization.

Control pattern

Skill graphs need path-aware review. The manifest must record not only which tools exist, but which outputs can be consumed as authority by which later steps. A safe skill should not be able to accidentally mint permission for another skill.

Vocabulary note

The report-derived label skill composition risk is used when capability flow, trust lift, or authorization blur turns benign skills into an unsafe path.

Skill composition risk is the mixed-case label used for validation and reader indexing.