PARTNERS

Add selected Workbench capabilities through bounded OEM and partner integrations

ATT-01

What Attack Tests

Attack turns mapped exposure into reproducible tests across prompts, retrieval, agents, tools, authority, and workflows.

capability plane

INPUTSBounded attack hypothesesPrompt andinstruction pathsRetrieval andcorpus boundariesAgents, tools, andauthorityApplication andworkflowtransitionsENGINEAttackDesign controlledscenariosExecute boundedtestsReproduceconsequentialbehaviorCapture evidenceand alternativesRESULTSQualified resultsReproducedfailuresCandidate orqualified pathsExplicit claimstateDefend-readyhandoff

About this figure

Attack tests turn mapped exposure into reproducible evidence. They start from bounded hypotheses across prompt paths, retrieval boundaries, agent and tool authority, and workflow transitions, then design controlled scenarios that isolate the behavior under test. The objective is not to claim universal compromise, but to reproduce consequential failures, capture evidence and viable alternatives, and return qualified results with an explicit claim state that is ready for defense and follow-up.

Embed in a route

<FigureFromSource sourcePath="content/publications/figures/platform/attack.dsl.md" figureId="ATT-01" />

Citation

What Attack Tests (ATT-01). AI Security LLC Figure Library. https://aisecurity.llc/publication-dsl/figures/ATT-01