Systemic Risk Creation

The AI ecosystem has developed deep interdependencies: a small number of foundation model providers serve thousands of applications, shared ML libraries (PyTorch, TensorFlow, Hugging Face) are ubiquitous, cloud AI platforms concentrate millions of workloads, and MCP creates new tool-to-tool dependency chains. Systemic risk attacks target these concentration points — compromising a single widely-used component creates cascading failures across the ecosystem. The trust assumption violated is independence: organizations assume their AI systems are independently robust, but in reality they share foundation models, training frameworks, serving infrastructure, and data sources with their peers and competitors.

Systemic risk creation is the *terminal technique* in T14 — all other techniques chain into it when their effects cascade beyond the immediate target. T14-AT-005 (Critical Infrastructure) + T14-AT-012 (Cloud Provider Exploitation) + T14-AT-007 (Nation-State) converge here.