{"slug": "maisi-mathematical-ai-safety-institute", "title": "Maisi – Mathematical AI Safety Institute", "summary": "The Mathematical AI Safety Institute (MAISI), an independent non-profit, has launched to develop mathematical foundations for the safety of powerful AI systems, with a team including 2026 Fields Medalist Jacob Tsimerman as Scientific Director and 1998 Fields Medalist Timothy Gowers on its advisory panel. MAISI is hiring 10–30 faculty for its January 2027 semester and 30–100 for its September 2027 Special Year, aiming to bring more mathematicians into AI safety research.", "body_md": "Mathematical AI Safety Institute\n\n# MAISI\n\nAn independent non-profit developing mathematical foundations for the safety of powerful AI systems.\n\n## Team\n\n- Scientific Director\n  - Jacob Tsimerman2026 Fields Medalist, and Professor at the University of Toronto. Jacob has received the 2023 Ostrowski Prize, the 2022 New Horizons in Mathematics Prize, the 2019 Coxeter–James Prize and the 2015 SASTRA Ramanujan Prize.\n- Scientific Advisory Panel\n  - Ravi VakilThe Robert Grimmett Professor of Mathematics at Stanford and, since 2025, President of the American Mathematical Society. Ravi has received the Presidential Early Career Award for Scientists and Engineers, is series editor of Springer’s Graduate Texts in Mathematics, and co-founded MathOverflow and the Proof School.\n  - Timothy Gowers1998 Fields Medalist, and the Chaire de Combinatoire at the Collège de France. Timothy is a life fellow of Trinity College, Cambridge. He received the 2016 Sylvester Medal, was knighted in 2012, was elected to the Royal Society in 1999, and received the 1995 Prize of the European Mathematical Society.\n  - Geoffrey IrvingCofounder and Chief Scientist of Resolution, formerly Chief Scientist of the UK AI Security Institute. Geoffrey led alignment teams at DeepMind and OpenAI, co-led Google Brain’s N2Formal neural theorem proving, and worked on computational physics and geometry at D. E. Shaw Research and Pixar. He holds a PhD in computer science from Stanford.\n  - Paul ChristianoDirector of the Alignment Research Center, and advisor to the Center for AI Standards and Innovation. Paul formerly led language model alignment at OpenAI, was head of safety at the US AI Safety Institute, and was an initial trustee of Anthropic’s Long-Term Benefit Trust. In 2023 he joined the UK’s Frontier AI Taskforce advisory board, and was among TIME’s 100 Most Influential People in AI.\n- Executive Director\n  - Andrew CritchHolds a 2013 Ph.D. in mathematics from UC Berkeley, applying algebraic geometry to machine learning models. Andrew spent five years as a research scientist at Berkeley’s Center for Human-Compatible AI, and continues there part-time. Prior to MAISI he co-founded theMultiplicity.ai, a platform for AI models supervising and correcting each other, and the Survival and Flourishing Fund.\n- Board of Directors\n  - Arul ShankarProfessor of Mathematics at the University of Toronto. Arul’s research is in arithmetic statistics, and joint with Manjul Bhargava, he has proved that the average rank of elliptic curves is finite, and that a positive proportion of elliptic curves satisfy the Birch and Swinnerton-Dyer conjecture. He is the past recipient of the Sloan Foundation Fellowship and the Simons Fellowship.\n  - Yevgeny LiokumovichAssociate Professor of Mathematics at the University of Toronto, where he leads the Fields–PrincInt Mathematics for AI Safety program and sits on the Fields Centre for Mathematical AI steering committee. Yevgeny works in geometric analysis, quantitative topology, and the mathematical foundations of AI alignment. With Fernando Codá Marques and André Neves he proved Gromov’s Weyl law conjecture, which won the 2022 André Aisenstadt Prize.\n  - Matthew ChouChief Executive Officer of Unipetro Group, a transportation infrastructure and lubricants business in Mississauga, and Petro-Canada Lubricants’ first international distributor since 1989. Matthew oversees the group’s lubricant sales to East Asian OEMs and its development portfolio. He holds graduate degrees from Columbia University and the University of St. Andrews, and previously built the Affordable Care Act team at Acumen, LLC.\n + Jacob Tsimerman and Andrew Critch\n\n## Apply to\n\nJoin MAISI\n\nWe’re hiring mathematicians to relocate to the Bay Area and begin researching AI safety ASAP.\n\nSpecifically we’re seeking 10–30 faculty for our January 2027 semester, and 30–100 for our September 2027 Special Year.\n\n[Apply](/apply)\n\n## Goals and\n\nStrategy\n\n### AI safety needs more mathematicians\n\nEvery time a new AI technology is developed or deployed, in some sense we’re rolling the dice with our future. In good ways and bad, AI is already disrupting the global economy. Today’s top AI experts even consider it an [extinction risk](https://aistatement.com/work/statement-on-ai-extinction-risk). Indeed, if we choose to build a self-sufficient artificial species that is more intelligent than humanity, it might compete with us for resources, and could plausibly replace us entirely.\n\nBut how big is the risk, really? Can it be measured and reduced? Subjective risk assessments from experts and public figures abound. While these estimates are an important expression of individual fears and uncertainties, humanity deserves a more principled measure of the risks at hand. A central reason the safety of powerful AI systems remains in question is that we lack a rigorous understanding of what it would mean to be safe, even in theory.\n\nEnter professional mathematicians. The foundations of statistics, physics, and even computer science are mathematical, giving us precise, robust, and reusable methods for calculating valid conclusions from valid premises. Now, we need similar mathematical foundations for AI safety, not only to measure risk, but to mitigate it.\n\n### Necessity is not sufficiency\n\nMathematical insights will be no panacea. Many ongoing AI safety efforts are still needed, including regulatory frameworks, hardenings of our social and technological infrastructure, and training techniques to make AI systems themselves more safety-oriented. Risks can arise in many ways, from many facets of society, and thus a multifaceted approach to safety is required.\n\nOur hope is not that mathematical solutions will somehow obviate other AI safety efforts. Rather, we hope to bolster them with more precise and rigorous definitions, measurements, and solution concepts, like the mathematics we use in physics and statistics to measure, discuss, and avert risks in nuclear science and epidemiology.\n\n### Key questions\n\nWhat premises would allow us to deduce with high confidence that a new AI technology will not cause a catastrophe?\n\nOver the years, various attempts have been made to formalize AI safety problems mathematically. Bits of progress have been made, but the issue of extinction-level risks from AI, or AI x-risk, remains unresolved.\n\nMathematically principled answers to these questions are waiting to be discovered. We have not, as yet, brought the full force of even a substantial fraction of the global mathematics community to contend with AI safety. At MAISI, we want to change that. Humanity deserves a mathematically principled account of existential safety for new AI technologies, on par with the mathematics that physicists use to ensure nuclear reactors are safe, or that statisticians use to track pandemics.\n\nWith that said, the core problems of AI x-risk as we face them today may turn out to be provably difficult or otherwise intractable, such that the answers we find could be negative. To name a few possibilities:\n\n1. Perhaps the current paradigm of AI development cannot be made existentially safe for humanity, and would thus require a major paradigm shift of some kind.\n2. Perhaps current or other AI paradigms could be existentially safe to build and use, but setting up the right configuration is provably chaotic and sensitive to noise in ways that are currently infeasible to address.\n3. Perhaps current or other AI paradigms could be existentially safe to build and use, but the task of verifying that fact has a provably high computational complexity class.\n\nIf any of these obstructions exists, we would like to know, and would like to support the mathematics community in pursuing that knowledge. Indeed, mathematics is one of the only disciplines that can prove one of its own problems to be unsolvable from a given foundation. In particular, if a given AI technology can never be made safe for some reason, perhaps a mathematical account of that impossibility can be brought to light before it’s too late.\n\n## Structure\n\nof MAISI\n\nMAISI will function first and foremost as a mathematical research institute, akin to the Institute for Advanced Study in Princeton or the Fields Institute in Toronto. Mathematicians will become Members of the institute for one or two semesters at a time, with the possibility of renewal. There will be offices surrounding a shared space for interacting and sharing ideas, and seminar rooms for internal presentations and invited speakers.\n\nVery roughly speaking, we will facilitate two kinds of discourse:\n\n**↔** Mathematicians\n\n**↔** Mathematicians\n\nThe first involves accomplished mathematicians discussing AI safety problems amongst each other, to become acquainted with existing literature and problems in the field. These conversations are where the incumbent wisdom and customs of the mathematics community will be felt the strongest.\n\nThe second involves accomplished experts in AI safety presenting existing solutions and open problems in the field, with mathematicians as the target audience. These conversations will ground theory in practice, seed new collaborations, and encourage Institute members to spend time working with and within our partner institutions in AI safety.\n\nWe expect that as the Institute takes shape, new AI safety research directions and perspectives will rapidly develop, and researchers will be encouraged to share these approaches to safety as quickly as possible. Unlike in standard academic mathematics, where results are carefully refined into papers and published over a matter of years, MAISI aims to move much more quickly, sharing ideas in the earliest form that might be of benefit to AI safety as a field. In particular, living documents with developing safety ideas will be shared long before they are finalized in a form suitable for mathematics journals.\n\n## Contact\n\nFor press inquiries, please use [press@maisi.org](mailto:press@maisi.org). For all other inquiries, please use [contact@maisi.org](mailto:contact@maisi.org). As the institute expands, additional email addresses may be made available on a per-topic basis.\n\n## Research\n\nCheck out our growing list of research directions — mathematically rich problems in AI safety, written for mathematicians.\n\n[Explore research directions](/research)", "url": "https://wpnews.pro/news/maisi-mathematical-ai-safety-institute", "canonical_source": "https://maisi.org", "published_at": "2026-09-09 05:00:15+00:00", "updated_at": "2026-09-09 05:20:33.414220+00:00", "lang": "en", "topics": ["ai-safety", "ai-research"], "entities": ["Mathematical AI Safety Institute", "Jacob Tsimerman", "Timothy Gowers", "Ravi Vakil", "Geoffrey Irving", "Paul Christiano", "Andrew Critch", "University of Toronto"], "alternates": {"html": "https://wpnews.pro/news/maisi-mathematical-ai-safety-institute", "markdown": "https://wpnews.pro/news/maisi-mathematical-ai-safety-institute.md", "text": "https://wpnews.pro/news/maisi-mathematical-ai-safety-institute.txt", "jsonld": "https://wpnews.pro/news/maisi-mathematical-ai-safety-institute.jsonld"}}