Quantitative research compares AI liability, slowdown, and pre-release regimes

Two sets of quantitative AI governance findings, both from the Forecasting Research Institute, reached related but distinct conclusions. The Leap panel found a US-only strict liability regime for AI outperforms a US-only slowdown or pre-release authorization regime, and is competitive with globally coordinated versions of those policies. A separate FRI analysis found that superforecasters strongly support an international body with pre-release authorization power over frontier AI models, and view US-only policies as reducing AI risk but less so than bilateral international policy. Experts surveyed also strongly oppose federal preemption of state AI laws and consider its passage risk-increasing. Despite these policy preferences, the same forecasters assign roughly 5.2% probability to 10% of humans dying from AI by 2050.