The PRISM AI Safety Research Fellowship is a highly structured, 16-week intensive program designed to produce high-impact, peer-reviewed research in critical areas of AI safety. The fellowship focuses on four primary pillars: mechanistic interpretability, AI governance, robust evaluation, and alignment. Unlike traditional research programs, PRISM utilizes a unique “blind peer voting” system where participants collectively vote on the most promising methodologies and research directions, ensuring a high-accountability and community-driven research environment. The program is built for researchers who are ready to contribute to the global AI safety landscape. Participants are expected to produce work suitable for submission to top-tier conferences such as NeurIPS, ICASSP, or specialized AI safety workshops. The fellowship provides a rigorous framework that includes weekly milestones, peer review sessions, and direct engagement with mentors from leading institutions like the OECD and major AI labs. This structure is designed to bridge the gap between theoretical knowledge and published research, making it an ideal platform for those looking to establish a serious track record in the field. Target Audience: STEM graduates, doctoral candidates, and independent researchers with a strong technical background in mathematics, computer science, or physics. The program also welcomes individuals from non-traditional pathways who can demonstrate exceptional technical proficiency and a commitment to AI safety. Geographic Eligibility: Remote/Global. The program is designed to be accessible to researchers worldwide, with synchronous sessions scheduled to accommodate multiple time zones. Tangible Benefits: Direct mentorship from established AI safety experts, opportunities for peer-reviewed publication in top-tier venues, access to a high-accountability research community, and a prestigious addition to a professional research portfolio. Application Deadline: June 15, 2026 (for the Summer 2026 cohort) Program Timeline: 16 weeks (June – September 2026)