LawZero: Yoshua Bengio's Nonprofit Scientist AI Safety Lab

A research organization for predictive AI

Yoshua Bengio publicly launched LawZero on June 3, 2025 as a nonprofit AI safety research organization. LawZero's biography of Bengio places its foundation in 2025; the earlier start of his research direction is not the organization's founding date. Its stated purpose is to develop safety-focused technical approaches rather than prioritize commercial deployment. That is an institutional aim, not proof that research is free of every external pressure. [1][2]

Scientist AI is the organization's proposed predictive approach: a system intended to understand and assess possibilities without adopting goals about which outcome should occur. LawZero's research page summarizes a July 2, 2026 formal safety argument, including assumptions about training dynamics and the rarity of dangerous predictors. It also discusses a guardrailed predictor that withholds high-harm predictions. These are research arguments and design goals, rather than demonstrated guarantees for deployed AI systems. [3]