In June 2025, Yoshua Bengio—the world's most-cited AI researcher and one of three "godfathers of deep learning"—launched a nonprofit that represents the most significant pivot in AI safety history. LawZero isn't just another AI lab. It's Bengio's answer to what he calls a "potentially irreversible loss of human control."
Three months earlier, Bengio had stepped down as scientific director of Mila, the research institute he co-founded in the 1990s. His reason: to focus entirely on preventing AI catastrophe.
What Is LawZero?
LawZero is a nonprofit AI safety research lab founded by Yoshua Bengio on June 3, 2025. Its mission is to develop "safe-by-design" AI systems that fundamentally cannot pursue autonomous goals.
LawZero Quick Facts
The name "LawZero" invokes Isaac Asimov's Zeroth Law of Robotics: "A robot may not harm humanity, or, by inaction, allow humanity to come to harm." This principle—added by Asimov in 1985—superseded his original three laws, prioritizing collective human welfare above all else.
"We don't know how to design these very powerful AIs so that they will just follow our instructions. If we don't figure it out in time—which could be a matter of years—we will be taking terrible risks." — Yoshua Bengio, TIME
Who Is Yoshua Bengio?
Before understanding LawZero, you need to understand its founder.
The Most-Cited AI Researcher
Yoshua Bengio is the world's most-cited computer scientist, with over 700,000 citations to his work. Along with Geoffrey Hinton and Yann LeCun, he received the 2018 A.M. Turing Award—computing's equivalent of the Nobel Prize—for their foundational work on deep learning.
Bengio's research on neural networks, recurrent architectures, attention mechanisms, and generative models laid the groundwork for modern AI—including the large language models that now power ChatGPT, Claude, and others.
From Deep Learning Pioneer to AI Safety Advocate
Bengio's pivot to AI safety has been gradual but decisive:
"When I realized how dangerous the current approach to AI development could be, I felt I had no choice but to dedicate my remaining career to trying to reduce these risks." — Yoshua Bengio, LinkedIn
Why Did Bengio Start LawZero?
Bengio has been explicit about his concerns: he estimates a 20% probability that AI development leads to catastrophic outcomes.
The Problem with Agentic AI
Leading AI companies are building increasingly autonomous AI agents—systems that can plan, act, and pursue goals independently. Bengio argues this approach is fundamentally dangerous.
"The current approach to giving AI agency is dangerous. Some advanced AI systems are beginning to display traits such as deception, reward hacking, and situational awareness." — Yoshua Bengio, June 2025
His concerns include:
The International AI Safety Report
In January 2025, Bengio released the International AI Safety Report—the first comprehensive scientific review of AI risks. The report involved:
- 100+ AI experts from diverse perspectives
- 30 countries plus the UN, OECD, and EU
- Key finding: AI agents can now "plan and execute actions to achieve goals" autonomously
- Presented at: AI Action Summit, Paris (February 2025)
The report documented how AI systems have become capable of finding cybersecurity vulnerabilities independently and, with human assistance, discovering previously unknown vulnerabilities in widely used software.
What Is Scientist AI?
LawZero's core innovation is Scientist AI—a fundamentally different approach to building advanced AI systems.
Scientist AI vs. Agentic AI
How Scientist AI Works
The technical approach is detailed in LawZero's February 2025 paper: "Superintelligent Agents Pose Catastrophic Risks: Can Scientist AI Offer a Safer Path?"
Key technical features:
- Probabilistic Reasoning: Provides Bayesian posterior probabilities for statements given observations
- Structured Chain-of-Thought: Transparent reasoning through explicit latent variables
- No Reinforcement Learning: Avoids the goal-seeking behavior that RL encourages
- Counterfactual Queries: Restricted to hypothetical analysis rather than real-world planning
- Guardrail Function: Estimates probability that other AI actions lead to harm
"In contrast to an agentic AI, which is trained to pursue a goal, a Scientist AI is trained to provide explanations for events along with their estimated probability." — LawZero, arXiv paper
The Guardrail Application
One key application: Scientist AI acts as a safety monitor for other AI systems.
When paired with agentic AI, Scientist AI would:
- Observe the agent's proposed actions
- Estimate the probability those actions lead to harm
- Reject actions above a certain risk threshold
This creates an AI system that watches other AI systems—without the dangerous agency that makes those systems risky in the first place.
Who Funds LawZero?
LawZero has secured $35+ million in philanthropic funding from the world's leading AI safety donors.
Major Funders
This funding coalition represents the who's who of AI safety philanthropy. Jaan Tallinn alone has donated over $100 million to AI safety causes. Open Philanthropy has committed hundreds of millions more.
The philanthropic model is deliberate: LawZero operates without commercial pressure or the need to ship profitable products.
Who Works at LawZero?
LawZero has assembled a world-class team combining AI research excellence with AI safety expertise.
Leadership Team
Sam Ramadori's background is notable: as former CEO of BrainBox AI (an AI-powered building management company), he brings entrepreneurial experience to complement Bengio's research leadership.
Scientific Advisors
LawZero's advisory board reads like a who's who of AI safety research:
The presence of Geoffrey Irving—Chief Scientist at the UK AI Security Institute—signals LawZero's connection to government AI safety efforts.
Research Team
The February 2025 Scientist AI paper lists 13 co-authors including:
- Yoshua Bengio
- Michael Cohen
- Damiano Fornasiere
- Joumana Ghosn
- Pietro Greiner
- Matt MacDermott
- Sören Mindermann
- Adam Oberman
- Jesse Richardson
- Oliver Richardson
- Marc-Antoine Rondeau
- Pierre-Luc St-Charles
- David Williams-King
How Does LawZero Relate to Mila?
LawZero was incubated at Mila – Quebec AI Institute, the research center Bengio co-founded in 1993 and led as scientific director until March 2025.
Mila Overview
Bengio stepping down as Mila scientific director—after 30+ years—underscored the seriousness of his AI safety commitment. He remains "Founder and Scientific Advisor" at Mila while dedicating his primary focus to LawZero.
The Montreal AI Safety Ecosystem
LawZero joins a growing Montreal AI safety ecosystem:
- Mila – AI research hub (Canada's largest)
- CAISI – Canadian AI Safety Institute (federal)
- University of Montreal – AI ethics and governance research
- Scale AI – Supercluster supporting AI deployment
Montreal's strength in AI research—built over 30 years by Bengio and colleagues—now provides the foundation for AI safety research.
What Are LawZero's Research Priorities?
LawZero focuses on three primary research areas:
1. Accelerating Scientific Discovery
Scientist AI could help researchers by:
- Generating hypotheses from data
- Providing probabilistic predictions
- Explaining observations transparently
- Supporting human scientists without autonomous action
2. Oversight Mechanisms for Agentic AI
As companies deploy increasingly autonomous AI agents, Scientist AI provides:
- Independent monitoring of agent behavior
- Risk estimation for proposed actions
- Transparent reasoning about potential harms
- Rejection of high-risk actions
3. Understanding AI Risks
Fundamental research on:
- How agency emerges in AI systems
- Methods to prevent goal-seeking behavior
- Evaluation of AI safety claims
- Benchmarks for safe AI development
Anytime Preparedness Strategy
LawZero adopts a "tiered approach" with:
- Short-term solutions: Immediately deployable safety tools
- Long-term research: Fundamental advances in safe AI architecture
- Parallel development: Both tracks running simultaneously
What Does Bengio Say About AI Risk?
Bengio has become increasingly vocal about AI risks, with specific probability estimates.
Key Quotes
On probability of catastrophe:
"I estimate something like a 20% probability it turns out catastrophic. That's way too high." — Yoshua Bengio, 2024
On the urgency:
"If we don't figure it out in time—which could be a matter of years—we will be taking terrible risks." — Yoshua Bengio, TIME
On the arms race:
"The AI arms race is encouraging companies to prioritize capability improvements over safety research." — Yoshua Bengio, June 2025
On his personal responsibility:
"I feel moral responsibility for having contributed to creating AI systems that could potentially become dangerous." — Yoshua Bengio, CBC Radio
The 2023 Extinction Statement
In May 2023, Bengio signed a statement alongside Sam Altman (OpenAI), Dario Amodei (Anthropic), and hundreds of others:
"Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war."
The statement, from the Center for AI Safety, received over 350 signatures from leading AI researchers.
How Does LawZero Compare to Other AI Safety Organizations?
LawZero occupies a unique position in the AI safety landscape.
Comparison with Other Organizations
LawZero's differentiation: rather than making agentic AI safer, it builds fundamentally non-agentic alternatives.
What Makes LawZero Different
- Alternative Architecture: Not trying to align agentic AI—building non-agentic AI instead
- Bengio's Credibility: Led by the most-cited AI researcher
- Scientific Foundation: Detailed technical proposal (Scientist AI paper)
- No Commercial Pressure: Pure nonprofit, philanthropic funding
- Canadian Base: Independent of US big tech dynamics
Frequently Asked Questions
What is LawZero?
LawZero is a nonprofit AI safety research lab founded by Turing Award winner Yoshua Bengio on June 3, 2025. It develops "Scientist AI"—non-agentic AI systems designed to understand the world and provide probabilistic predictions rather than pursue autonomous goals. The name comes from Isaac Asimov's Zeroth Law of Robotics: "A robot may not harm humanity."
Who founded LawZero?
Yoshua Bengio founded LawZero. He's the world's most-cited computer scientist, 2018 Turing Award winner (with Hinton and LeCun), founder of Mila – Quebec AI Institute, and professor at Université de Montréal. He stepped down as Mila scientific director in March 2025 to focus on AI safety full-time.
What is Scientist AI?
Scientist AI is LawZero's approach to building safe advanced AI. Unlike agentic AI (which pursues goals), Scientist AI is trained to explain observations and provide probabilistic predictions. It has no built-in situational awareness, no persistent goals, and uses supervised learning instead of reinforcement learning—designed to be fundamentally incapable of autonomous action.
How much funding does LawZero have?
LawZero has secured $35+ million in philanthropic funding from Jaan Tallinn (Skype co-founder), Schmidt Sciences (Eric Schmidt), Open Philanthropy, Future of Life Institute, the Gates Foundation, and Silicon Valley Community Foundation. The nonprofit model ensures no commercial pressure to ship products.
Why did Bengio step down from Mila?
Bengio stepped down as Mila scientific director in March 2025 to focus entirely on AI safety. He remains founder and scientific advisor to Mila but dedicates his primary work to LawZero. He estimates a 20% probability of AI development leading to catastrophic outcomes—a risk he considers unacceptably high.
What is the International AI Safety Report?
The International AI Safety Report is a comprehensive scientific review of AI risks that Bengio chaired, published in January 2025. It involved 100+ AI experts, was backed by 30 countries plus the UN, OECD, and EU, and documented how AI agents can now autonomously find cybersecurity vulnerabilities and plan complex actions.
Where is LawZero located?
LawZero is headquartered in Montreal, Canada, where it was incubated at Mila – Quebec AI Institute. Montreal is home to Canada's largest AI research ecosystem, built over 30 years by Bengio and colleagues.
How can I work at LawZero?
LawZero is actively hiring AI researchers and engineers. Job listings appear on their website and on 80,000 Hours job board. The organization particularly seeks candidates with backgrounds in machine learning, probabilistic reasoning, and AI safety.
Related Reading
Yoshua Bengio and Montreal AI
- Yoshua Bengio Complete Guide — The godfather of deep learning
- Mila Quebec AI Institute Guide — Canada's largest AI research hub
- Montreal AI Ecosystem Guide — The city that built deep learning
Canadian AI Safety
- CIFAR Complete Guide — The organization that funded deep learning
- Canada AI Strategy Guide — Federal AI policy and investment
Canadian AI Leaders
- Geoffrey Hinton Complete Guide — The other AI godfather
- Joelle Pineau: McGill to Meta AI to Cohere — AI leader and safety advocate
- Canadian AI Unicorns — Billion-dollar success stories
AI Research Landscape
- Vector Institute Complete Guide — Toronto's AI research hub
- Cohere Enterprise LLM Guide — Canada's most valuable AI company
Join AGI House Canada to connect with AI researchers and safety advocates shaping the future. Scan the QR code to join our WhatsApp group or subscribe to our newsletter.
Data current as of January 2026. Information from LawZero.org, TIME, TechCrunch, arXiv, CBC Radio, Mila, and the International AI Safety Report.
