Venue Opens for Networking
Join us in WorkAdventure to meet SPAR researchers and network with others interested in AI safety.
May 23, 2026
100+ AI Safety Research Projects
SPAR is a research mentorship program helping aspiring researchers contribute meaningfully to frontier AI safety work. This round's virtual Demo Day (hosted in WorkAdventure) showcases 100+ technical and governance projects developed over 3 months by this round's stellar mentors and mentees. Projects from previous SPAR rounds have been accepted at ICML and NeurIPS, covered by TIME, and led to full-time job offers for mentees.
SPAR is run by Kairos, a nonprofit whose mission is to help society navigate the transition to transformative AI. Express interest for future iterations of SPAR here.
Registration for Demo Day is closed
Virtual Event
Date
May 23, 2026
Time
8:30 AM - 1:30 PM PT
Platform
WorkAdventure
Registration is closed
Projects
Career Fair
Connect with leading AI safety organizations, including:
May 23, 2026
Event Schedule
8:30 AM - 1:30 PM PT
Timezone:
Timeline
Poster Session
Come see the results of 100+ AI safety and policy research projects and discuss them with their teams.
Lightning Talks
Come hear five-minute presentations on SPAR's research projects.
AI Safety Conversations
Choose an AI safety discussion topic and meet with others interested in discussing it.
Career Fair (SPAR Participants Only)
Connect with leading AI safety organizations, network with recruiters, and learn about career opportunities in AI safety research and policy.
Career Fair (Open to All)
Connect with leading AI safety organizations, network with recruiters, and learn about career opportunities in AI safety research and policy.
Awards and Closing
Wrap up the day with award presentations and closing remarks.
Awards
Spring 2026 Winners
Best Poster
$1,000 / member
+ $500 / member for 2nd place
1st Place
Out of Context Obfuscation: What Facts Matter for Probe Evasion?
William Soylemez, Fadi Benzaima, Faraz Ahmed Siddiqui
2nd Place
Interpreting Prefill Awareness
Andrew Fletcher
3rd Place
De-risking Interconnect Limits for AI Verification
Diana Sarbakysh, Art Moskvin
Honorable Mention
FragBench: Cross-Session Attacks Hidden in Benign-looking Fragments
Astha Mehta, Raymond Lee, Olivia McGoffin, Niruthiha Selvanayagam, Hengxu Li
Honorable Mention
AIs Hacking ML Training Tasks
Kaustubh Kislay, Leon Eshuijs, Meiri Anto, Pradyumna Shyama Prasad, Julian Moncarz
Honorable Mention
Language Models Can Recognize Dropout and Gaussian Noise
Spencer Kitts
Best Lightning Talk
$2,000
+ $1,000 for runner-up
1st Place
Identifying function-relevant, sequence-agnostic features in a protein model using sparse autoencoders
Isha Harris
2nd Place
Do LLMs have preferences?
Sam Wang
3rd Place
Not All Eval-Awareness Is Equal: Capabilities-Framing Predicts Compliance
Allison Zhuang