CASCADE: Cross-Agent Supply Chain Audit and Defense Ecosystem

CASCADE is a framework paper on agentic AI supply-chain risk. It models autonomous systems as dependency networks so teams can identify concentration, score attack-surface exposure, and reason about how upstream failures cascade downstream.

The paper's core claim is that model-level evaluation is not enough when agents rely on shared model providers, data pipelines, tooling layers, and compute infrastructure. Governance teams need a way to map chokepoints, simulate propagation, and design layered defenses before a dependency failure becomes a fleet-wide incident.