PRISM — Peer-vetted Research Initiative for Safety Methodologies — is a 16-week research fellowship run by Women Who Do Data (W2D2), a US nonprofit that gives underrepresented data and AI professionals a place to learn and lead. Fellows produce peer-reviewed research in interpretability, AI governance, evaluation and alignment.
The programme swaps mentor-led selection for peer selection. Teams pitch the literature, propose methods, then decide by anonymous vote which method the team will pursue, so scarce mentor time is spent where it cannot be replaced. The 2026 cohort is planned at 48 fellows across 12 research groups, each group a team of four attached to a named project lead.
PRISM was piloted inside EleutherAI’s Summer of Open AI Research (SOAR) with 50 fellows across five research tracks; that pilot produced four papers, two of them accepted at NeurIPS and ICASSP 2025. The 2026 cohort is supported by a grant from the Long-Term Future Fund (EA Funds). The programme director is Archana Vaidheeswaran.
Benefits & funding
- Sixteen weeks of structured research inside a team of four, in one of 12 research groups led by a named researcher.
- Research areas span technical safety, benchmarks and evaluations, governance, and frontier risks — for example multilingual safety evaluations, red-teaming protein foundation models, and applying epidemiological methods to AI harm monitoring.
- Where a team’s work is accepted at a conference, one team member is funded to present it.
- The 2026 cohort is backed by a grant from the Long-Term Future Fund (EA Funds).
- No participation fee is published, and no stipend for fellows is published either.
- The target output is a conference-ready paper; the 2025 pilot produced four papers, two accepted at NeurIPS and ICASSP 2025.
Eligibility
- Requirements are set per research project rather than by one central bar; each of the 12 projects names the background it wants — one, for example, asks for a basic grounding in computer science together with a public policy or economics background.
- No nationality or residency requirement is published.
- No degree level, age or gender requirement is published.
- Applicants may apply to several of the listed projects, or to only one or two.
- PRISM states its purpose as widening access for underrepresented talent in AI safety, and its host organisation serves underrepresented data and AI professionals; the public pages present this as the programme’s focus rather than as a bar on who may apply.
- The site does not publish a weekly time commitment.
Can you apply?
Open to all applicants
No nationality, residency, education, or experience limits are listed.
Circlebreaker is a free database of global fellowships, scholarships and programs — every one decoded like this.
We only check the criteria we have data for. Always confirm full eligibility on the official page before applying.
Application process
- 1Read the Summer 2026 project list and shortlist the projects that match your background.
- 2Complete the online application form linked from the PRISM homepage, choosing the projects you want to be considered for.
- 3Submit before the cycle closes — the 2026 round closed at 08:00 PDT on 25 May 2026 — then check for the confirmation email.
- 4Selected applicants join a team of four and start the 16-week programme.
Program timeline
25 May 2026
Applications for the 2026 cohort closed at 08:00 PDT
The 16-week programme runs with 48 fellows in 12 teams
Summer 2026
Week 2
Literature pitch sessions
Week 3
Methodology proposals
Week 4
Anonymous team vote on methodology
Week 5 onwards
Paired experiments with weekly presentations to teammates
Start and end dates for the cohort are not published on the site