Unbounded Harms, Bounded Law: Liability in the Age of Borderless AI

Current liability laws are geographically limited but AI harms are global: The paper identifies a fundamental mismatch—AI systems operate across borders and can cause harm internationally, yet legal responsibility frameworks remain tied to individual countries. This creates gaps where victims in one country may have no recourse when harmed by AI deployed from another.

We need to learn from how other high-risk industries handle cross-border liability: The authors examine successful models from nuclear power, vaccines, and financial systems to identify proven legal tools like "strict liability" (holding companies responsible without proving negligence) and "risk pooling" (spreading costs across an industry). These mechanisms could be adapted to ensure AI companies compensate victims regardless of where harms occur.

AI supply chains hide responsibility: Because AI systems involve multiple companies across different countries—data providers, model developers, deployers—it's often unclear who should be held accountable when something goes wrong. The paper argues this diffusion of responsibility is a structural problem that current laws don't adequately address.

Competitive pressure between nations may prevent effective global AI governance: The paper warns that countries racing to develop AI capabilities may resist strong international liability rules that could slow their industries, making cooperative solutions difficult to achieve and leaving vulnerable populations underprotected.

Preventing harm before it happens isn't enough: While the field has made progress identifying and preventing AI risks upfront, there's been insufficient attention to what happens after harm occurs—who pays, how victims get compensated, and how to ensure accountability.