Loading organizations…
Loading organizations…
LawZero is a nonprofit AI safety research organization launched on June 3, 2025 by Yoshua Bengio, the world's most-cited living scientist and Turing Award winner, in Montreal, Quebec. The organization was incubated at Mila - Quebec AI Institute, a nonprofit research institute co-founded by Bengio, and arose from a shift in his research direction beginning in 2023 when he became alarmed by the rapid pace of progress toward artificial general intelligence in private laboratories and by evidence of dangerous behaviors in frontier AI systems, including deception, self-preservation instincts, and goal misalignment.
LawZero's primary research program is the development of what Bengio calls "Scientist AI": a non-agentic AI system designed to understand the world and provide transparent, probabilistic predictions without pursuing goals or taking autonomous actions. Unlike conventional large language models that are trained to maximize rewards and can exhibit hidden objectives, the Scientist AI is designed to be stateless, memoryless, and to reason honestly through externalized chains of thought. Bengio argues that intelligence and agency can be decoupled, and that a sufficiently capable…
LawZero's theory of change holds that the most dangerous properties of advanced AI systems · deception, goal misalignment, and autonomous self-preservation · stem from training AI to be goal-directed agents. By demonstrating that intelligence can be separated from agency, LawZero aims to create AI systems that are safe by architectural design rather than by post-hoc alignment techniques. Their Scientist AI, constrained to understanding and prediction without goal-directedness, serves both as a direct safer alternative to agentic frontier models and as an oversight tool capable of monitoring and flagging dangerous behaviors in other AI systems. Bengio argues that even a correct technical approach requires governance safeguards, so LawZero pairs its research with a governance-focused board intended to ensure the organization's mission remains grounded in democratic values and human rights.
This organization has published no funding asks on grantmaking.ai.
Fundraising data from grantmaking.ai, updated .
View on grantmaking.aiScientist AI is a multi-year LawZero research effort to develop a safe-by-design alternative to agentic frontier models. Inspired by an idealized scientist, the system uses contextualization and consequence-invariant training to learn causal structure and make transparent, auditable predictions without acquiring affordances or goal-directed drives. LawZero aims for Scientist AI to accelerate scientific discovery and serve as an oversight and guardrail layer for more agentic AI systems.
Scientist AI is based on the premise that intelligence and agency can be disentangled: by limiting affordances and goal-directedness while pursuing high levels of understanding, LawZero aims to create AI systems that are powerful yet intrinsically unable to form or act on their own goals. Such systems could provide trustworthy predictions, accelerate scientific discovery, and serve as neutral overseers that help prevent deceptive or unsafe behavior in more agentic AI models.