A participant-led unconference gathering the global AI safety community, online and at local sites, Nov 20-22, 2026.
A participant-led unconference gathering the global AI safety community, online and at local sites, Nov 20-22, 2026.
Project Details
Updated 07/13/26 · Provided via application · VerifiedThe AI Safety Unconference 2026 (formerly VAISU) is a participant-led event on Fri-Sun Nov 20-22 to bring together the global community working on AI existential and catastrophic risks.
It is a multidisciplinary event which will invite participants from all fields of AI safety. It is an online event, with coordination and a toolkit to organize in-person sites (especially in the network of AI safety coworking spaces) supported by microgrants. Participants provide sessions: talks, small workshops, roundtables, office hours, targeted feedback, etc. The schedule is computed just-in-time to optimize participant preferences and session host availability. It will feature an agenda red-teaming track, 1:1 networking time with a matchmaking recommender system, and a general poster session.
Previous AI safety unconferences were assembled via a mashup of tools. This edition we are building a custom application to provide one coherent system with a better UX. We are interested in pursuing the series over the coming years.
AISU 2026 is a project of the HΩ AI safety non-profit, with Orpheus Lummis (founder), Sarah Khalid (senior PM), and Vadym Sulzhenko (AI safety field-building) as core organizers. History: In 2024, Linda Linsefors, who had been organizing VAISU since 2019, and Orpheus who had been organizing AISU@NeurIPS since 2018, came together through an AI Safety Camp project to organize VAISU 2024 which had 400 registrations & 40 sessions. After that edition, Orpheus took the lead for future VAISUs.
The project will produce a proceedings website with curated notes and the results of the agenda red-teaming track, in addition to a playlist of videos. We aim at 1000 registrations, and over 50 sessions this year.
Theory of Impact
Updated 07/13/26 · By grantmaking.aiReducing AI x-risk depends not only on work inside frontier labs and governments but on the collective capacity of the field in general: whether talent finds the right problems, whether weak agendas get corrected, and whether researchers who disagree ever meet. AISU is an intervention on that.
Cross-paradigm convening exists in 2026 but is not sufficient: FAR AI's Alignment Workshop is invite-only and senior, Iliad is single-paradigm, EA Global is not dedicated to AI safety, MAISU was minimal. What is missing is a venue allowing dialogue across the fields of AI safety, and where newcomers and researchers outside the hub cities get contact with each other and senior folks.
The unconference format allows for the community to form sessions that it thinks are most relevant.
Matchmaking and supported in-person nodes create more durable ties and collaborations. The mentorship networking helps newcomers with their trajectories. This is how talent finds the right problems and collaborators.
The agenda red-teaming track invites researchers from multiple perspectives to produce written critiques which will be published in the proceedings. This will provide error correction, so less effort is wasted on agendas that don't survive scrutiny.
People
Updated 07/13/26 · By grantmaking.aiTeam Member
Funding Details
- Jul 1, 2026
- Nov 30, 2026
- -
- -
- -
- -
- -
- -
- -
- -
Great initiative with lots of local field-building potential