This article has been nominated for redirecting to Yoshua Bengio § Career and research. You are welcome to participate in the redirecting discussion, which will decide whether to redirect it to Yoshua Bengio § Career and research. This discussion may also result in the article being deleted, merged, or draftified. |
LawZero is a Canadian nonprofit artificial intelligence safety research organization based in Montréal, Quebec. It was founded on 3 June 2025 by Yoshua Bengio and incubated at Mila – Quebec AI Institute, which remains its operating partner.[1][2][3]
| Formation | 3 June 2025 |
|---|---|
| Founder | Yoshua Bengio |
| Type | Nonprofit research organization |
| Focus | AI safety |
| Location | |
Key people |
|
| Affiliations | Mila – Quebec AI Institute (operating partner) |
| Employees | ~30 (2026) |
| Website | lawzero |
LawZero's principal research direction, which it calls "Scientist AI", aims to develop a non-agentic AI system that would understand and make probabilistic predictions about the world without holding goals or taking independent actions of its own.[1][4][5][6] The organization has stated that its aim is to advance "safe-by-design" AI systems through research it positions as insulated from commercial and government pressures.[1][7]
At launch, LawZero raised approximately US$30 million in philanthropic funding; Time reported named donors including Jaan Tallinn, Schmidt Sciences, Coefficient Giving and the Future of Life Institute.[1][8] According to LawZero, a subsequent grant from the Gates Foundation in August 2025 brought total philanthropic funding to more than US$35 million.[9] In February 2026, The Globe and Mail reported that the Government of Canada had signed a letter of intent with LawZero for further financial backing, with figures of more than $100 million reportedly under discussion.[2] By February 2026, the organization employed approximately 30 people and had stated plans to expand to more than 100.[2]
History
editLawZero was launched on 3 June 2025 in Montréal as a nonprofit organization incubated at Mila – Quebec AI Institute.[1][10][3] Yoshua Bengio, who took the role of co-president and scientific director, has attributed the project's origins to a change in his research direction beginning in 2023, prompted by his view that the pace of progress by private AI laboratories toward artificial general intelligence warranted dedicated work on AI safety techniques separable from commercial development.[1]
In an interview with the Financial Times coinciding with the launch, Bengio said that current foundational AI models had begun to display behaviours including lying, cheating and self-preservation strategies, and presented LawZero's research as a response to those tendencies.[7][11] At launch, the organization had a team of more than 15 researchers.[1][4]
Mission and approach
editLawZero describes its mission as the development of AI systems that are "safe by design" through technical research that it positions as insulated from market and government pressure.[1][12] Time has reported the organization as pursuing a path that Bengio characterized as "fundamentally different" from that of major AI laboratories, with the explicit aim of producing AI that can act as a guardrail rather than as an autonomous agent.[1] LawZero hosts a research stream at the MATS AI-alignment training program for summer 2026.[13]
Scientist AI
editLawZero's principal research direction is a class of system Bengio calls Scientist AI. As described by the organization and reported by various media outlets, the system is intended to be non-agentic (focused on understanding and prediction rather than autonomous action); to produce explicit, externalized reasoning for its outputs; and to express calibrated uncertainty about its own answers.[1][4][5]
A 2025 arXiv preprint led by Bengio argues for the approach by reference to what its authors characterize as catastrophic risks from highly capable agentic systems, proposing Scientist AI as a possible "safer path".[6] In a February 2026 white paper, LawZero describes the design in terms of three mechanisms: contextualization, meaning the transformation of training data so that factual claims are disentangled from opinions; consequence invariance, preventing the system from being optimized on the basis of downstream real-world outcomes; and a generator–estimator architecture in which a creative generator is held accountable by a neutral estimator.[14]
As noted, one envisaged use of Scientist AI is as a guardrail for agentic AI systems, with the non-agentic system predicting whether a proposed action by an agentic AI is dangerous and, if the probability of harm exceeds a threshold, blocking it.[1][4]
Leadership and governance
editBengio serves as co-president and scientific director.[2][1] Sam Ramadori is co-president and executive director.[9] By February 2026 the organization employed approximately 30 people and, The Globe and Mail reported, had stated plans to expand to more than 100 within a year.[2]
In January 2026, LawZero announced its inaugural board and global advisory council. The board is chaired by Maria Eitel, founder of the Nike Foundation and of Girl Effect, and includes Mariano-Florentino Cuéllar (president of the Carnegie Endowment for International Peace and a former California Supreme Court justice), historian Yuval Noah Harari, Valérie Pisano (president and chief executive of Mila), Sir John Rose (former chief executive of Rolls-Royce) and Bengio. The global advisory council includes Stefan Löfven, former Prime Minister of Sweden.[15]
Funding
editLawZero raised approximately US$30 million in philanthropic funding prior to its public launch in June 2025; Bengio told Axios that the amount was expected to support the basic research effort for approximately 18 months.[12][1] Several of the initial donors — including Coefficient Giving, the Future of Life Institute and Jaan Tallinn — have previously funded other efforts to reduce existential-risk.
In August 2025, LawZero announced a grant from the Gates Foundation to support its work on AI safety benchmarks, algorithms, and applications to scientific research and drug discovery; according to LawZero, the grant brought total philanthropic funding raised since launch to more than US$35 million.[9]
In February 2026, The Globe and Mail reported that the federal government had signed a letter of intent with LawZero for further financial backing, with a figure of more than $100 million reportedly under discussion. Federal AI minister Evan Solomon described the amount as "substantial" and framed the commitment as part of a broader Canadian regulatory, legislative and technical strategy for fostering trust in AI. The paper noted that Ottawa had previously committed approximately $240 million in support of Cohere, a Toronto-based AI company. The same report quoted LawZero stating that "the vast majority [of expenditure] is for compute, because we aim to develop technology that requires a lot of compute".[2]
References
edit- 1 2 3 4 5 6 7 8 9 10 11 12 13 Henshall, Will (June 2025). "The Most-Cited Computer Scientist Plans to Make AI More Honest". Time. Retrieved 16 May 2026.
- 1 2 3 4 5 6 Castaldo, Joe (19 February 2026). "Ottawa plans major investment in AI pioneer's non-profit". The Globe and Mail. Retrieved 16 May 2026.
- 1 2 "Yoshua Bengio Launches LawZero: A New Nonprofit Advancing Safe-by-Design AI" (Press release). LawZero. 3 June 2025. Retrieved 16 May 2026.
- 1 2 3 4 "AI pioneer announces non-profit to develop 'honest' artificial intelligence". The Guardian. 3 June 2025. Retrieved 16 May 2026.
- 1 2 Samuel, Sigal (19 June 2025). "He's the godfather of AI. Now, he has a plan to keep us safe from it". Vox.
- 1 2 Bengio, Yoshua; et al. (24 February 2025). "Superintelligent Agents Pose Catastrophic Risks: Can Scientist AI Offer a Safer Path?". arXiv:2502.15657 [cs.AI].
- 1 2 "'Godfather' of AI Yoshua Bengio says latest models lie to users". Financial Times. 3 June 2025.
- ↑ "AI Pioneer Bengio Launches Research Group to Build Safer Agents". Bloomberg News. 3 June 2025. Retrieved 16 May 2026.
- 1 2 3 "LawZero Receives Grant to Develop Safe-by-Design AI Systems That Can Improve Scientific Discovery" (Press release). LawZero. 25 August 2025. Retrieved 16 May 2026.
- ↑ Katz, Leslie (4 June 2025). "'Godfather Of AI' Launches Nonprofit Focused On Safer Systems". Forbes.
- ↑ "AI godfather Yoshua Bengio says current AI models are showing dangerous behaviors like deception, cheating, and lying". Fortune. 3 June 2025. Retrieved 16 May 2026.
- 1 2 "Yoshua Bengio launches LawZero to rethink AI safety". Axios. 3 June 2025. Retrieved 16 May 2026.
- ↑ "LawZero at MATS: Summer 2026". ML Alignment & Theory Scholars Program. 2026. Retrieved 16 May 2026.
- ↑ LawZero (5 February 2026). The Scientist AI: Safe by Design, by Not Desiring (Report). LawZero. Retrieved 16 May 2026.
- ↑ "LawZero Appoints 7 Global Leaders, Including Top AI and Business Figures as well as a Former Head of Government, to its Board and Global Advisory Council" (Press release). LawZero. 15 January 2026. Retrieved 16 May 2026.