University groups are one of the most common routes into AI safety work: a 2023 survey of people working on catastrophic risks found university group involvement was the single most common career influence. Dedicated field-building organizations like Kairos exist largely to support these groups, and the ceiling for their success is high: groups like MAIA, AISST, and BASIS have grown from paper-reading groups into some of the field's strongest talent pipelines [2]. It is not uncommon for top talent to go from “vaguely interested student” to full-time hire in 6-12 months.
Kairos's Pathfinder fellowship does a great job of providing existing founders with mentorship, funding, and resources to ensure their groups are impactful. But Kairos doesn't yet have the capacity to systematically search for and reach out to good candidates for Pathfinder. This means that today, a group forms only when a rare student fits a narrow path:
Our central bet is that there already exist capable would-be founders at schools that lack a group. They are capable enough to start one, but have been deterred by the costs, aren't aware of the resources available to them, or have simply never considered the idea. Group conception currently relies largely on an ultra-rare student stepping up with little initial guidance, and while great support exists once this founder emerges, deliberately finding and convincing capable students to become group founders has seldom been done [5]. As a result, only 37 of the top 100 US universities have groups, and even in the top 30, there are some 200k students at a school without a group
To fix this gap in the student-to-successful-founder pipeline, we’re launching the AI Safety Seeding Initiative to search for latent founders, convince them to start a university group, and provide initial support to help them succeed in Pathfinder and beyond. This initiative is funded by and in partnership with Kairos as part of their ongoing efforts to strengthen and grow the infrastructure for AI safety talent.
We're a team of AI safety field-builders who have experience founding and directing university groups. Jason Chin (founder of VAISI) and Thomas Rodskog (founder of AI Safety at UCSB) are co-leading the project, with direct support from advisors Tzu Kit Chan (Chief of Staff, Atlas Computing; seeded/mentored 50+ university groups around the world, primarily in the US) and Jeremy Kintana (Generalist, Kairos; former director of WAISI).
1. If you're a student (or a potential founder), reach out. There may be no group at your school yet. Check your school and book a 20-minute intro call. No prep needed.
2. Refer a potential founder (our biggest ask). If you know a sharp student interested in AI safety at any strong university without a university group, send them our way. Have a low bar. Even a low-probability referral is probably worth your time.
3. Refer a (non-AI-safety) student. If you know a student at one of the universities listed, also send them our way! Even if they’re unable or uninterested in being a founder, they can be equally useful in helping us find founders by connecting us with people they know or posting about it on campus or in listservs and online groups used by students.
**4. Help us with the AI Safety Seeding Initiative. **Suggest search strategies, share useful data on potential founders, or offer to work with our team. You can also share private or anonymous feedback with this form.
List of our priority schools:
Duke University
University of Pennsylvania
UNC Chapel Hill
Vanderbilt University
New York University (NYU) Johns Hopkins University
University of Notre Dame
Emory University
University of Southern California (USC) Texas A&M University
Ohio State University (OSU) University of California, Davis (UC Davis)
University of Georgia (UGA) William & Mary
Tulane University
Wake Forest University
Boston College (BC) Northeastern University
University of Miami (U Miami)
Virginia Tech (VTech)
North Carolina State (NC State)
Penn State University (PSU)
Michigan State University (MSU)
Brigham Young University (BYU)
Spelman University
Howard University
Harvey Mudd College
Pomona College
Claremont McKenna College
Swarthmore College
Williams College
Barnard College
Carleton College
Wellesley College
Bowdoin College
Grinnell College
Priority schools from manually compiled dataset: Duke, Penn, UNC, Vanderbilt, NYU, Johns Hopkins, Notre Dame, Emory, USC, Texas A&M, Ohio State, UC Davis, UGA, W&M, Tulane, Wake Forest, BC, Northeastern, U Miami, VTech, NC State, PSU, MSU, BYU, Spelman, Howard, Harvey Mudd, Pomona, CMC, Swarthmore, Williams, Barnard, Carleton, Wellesley, Bowdoin, Grinnell.
There are two main costs in founding:
-
Starting a new group is scary. You have to speak publicly, become "the AI safety person" on campus, and trust your own knowledge enough to lead, usually with no one beside you.
-
Organizing, especially early on, is a huge time commitment, and often the students who are best suited to start a group are busy with other ambitious goals.
Tzu seeding Caltech and posts like this.
See dataset in footnote 1.
The two main forms of support we're thinking of are: 1) strategic advising, and 2) offering to absorb organizing hours by acting as forward-deployed organizers (either ourselves or volunteers from other groups).