Instrument Choice in AI Governance: The Case for a Liability-Centered Framework

Abstract

With rapid advances in artificial intelligence, policymakers face an instrument-choice problem under deep uncertainty: whether to govern primarily through ex ante controls or ex post accountability. Ex ante regulatory strategies—licensing, prescriptive rules, approvals, and compute-based controls—force early commitments to contested forecasts about capabilities, harms, and precautions—commitments that must be revised as systems and deployment practices evolve.

This Article argues that AI governance should adopt a liability-centered framework, with strict liability, grounded in familiar U.S. tort concepts such as abnormally dangerous activities, as the central instrument for many AI-caused harms. Strict liability largely sidesteps epistemic disagreement by operating ex post: courts can assess responsibility after deployment, when system behavior, harm patterns, and feasible precautions are observable. That timing advantage is paired with an incentive advantage: by forcing developers and deployers to internalize the costs of harms without requiring proof of fault, strict liability creates strong incentives to invest in safety and to update precautions as the frontier shifts. Building on prior work, the Article explains how near-miss punitive damages for uninsurable tail risks and capability-scaled liability insurance can reinforce a liability-centered regime.

Through systematic comparison with licensing, approvals, prescriptive regulation, and compute governance, the Article then develops a framework for non-liability tools. It distinguishes supportive policies that make liability workable in practice (for example, traceability requirements such as transparency, information preservation, and incident reporting), complementary policies that address residual or non-compensable risks beyond liability’s reach, and substitutionary policies that serve as second-best substitutes when robust liability is blocked by legal or political constraints, including federal preemption or displacement of state law.