AI Supply Chains: An Emerging Ecosystem of AI Actors, Products, and Services

Information loss in AI supply chains creates real-world risks: The paper demonstrates that explanations and information about AI models become distorted as they pass between different actors in the supply chain, leading to misunderstandings about how AI systems actually work—a finding with concrete implications for how AI products are deployed and understood by end users.

Upstream decisions have cascading downstream effects: Design choices made by base model providers (companies creating foundational AI models) significantly impact all the AI products built on top of them, meaning responsibility and consequences are distributed across the entire supply chain rather than concentrated with final product developers.

Current AI regulation doesn't account for supply chain complexity: The paper shows that both AI development and regulation become substantially more complicated when multiple specialized actors are involved, suggesting that existing regulatory frameworks may not adequately address accountability and safety across fragmented supply chains.

AI supply chains reflect healthy industry specialization, but create new challenges: While the emergence of specialized AI actors (dataset providers, model creators, service integrators) mirrors normal economic growth in other industries, it introduces novel coordination and transparency problems that don't have clear precedents in other sectors.

We need better frameworks for understanding AI supply chain risks: By modeling supply chains as directed graphs, the paper provides a foundation for systematically studying how different network structures and actor relationships create different types of risks and accountability gaps.