Frontier Technologies, Complex Externalities, and the Evolution of Private Order: Implications for AI Governance

Abstract Artificial intelligence (AI) creates externalities whose sources and magnitude are difficult to know before deployment. Most of the work on AI governance asks how public authorities should regulate these risks. We ask how public regulation affects the discovery of private order. The cases of electrical safety certification, cyber insurance, and early radio broadcast illustrate how experience can reveal conflicts and support flexible private governance. A model of reversible deployment identifies two costs of an imposed public veto. By blocking deployment, it can prevent the learning needed to identify effective safeguards and allocate authority. Because affected parties cannot bargain around it, the veto can also continue to block mutually beneficial activity after they acquire more information. The cases and the model suggest that the costs of public regulation include its effects on private governance that might otherwise emerge. This consideration matters for the design of institutions that govern AI.