Project bank
Real problems sent in by people working in AI safety, plus open problems with no set contributor yet. Everyone in the cohort gets one project, cut down until it can be finished in five days and used afterwards.
Harry Waterman · Generator Residency · Fellow
Build a run-ready workshop that helps AI safety people improve their models of how the US government responds to crises. Participants get a historical crisis as it looked on day 0, predict which agencies will act, under what authorities, and on what timeline, then compare…
Matt Handzel · Kairos / Constellation · Generator Resident
Create a methodology (and the material behind it) that lets applicants work out whether they'd actually enjoy a given AI safety role before sinking effort into applications and work trials.
Sohan Venkatesh · LASR Labs · Research Fellow
A reusable toolkit that stress-tests any linear deception probe against the failure modes recent research has identified, but never packaged. Given a trained probe and a target model, it would run a standardized battery: an eight-style distributional shift test, cross-domain…
Alejandro Acelas · Impact Ops / 80,000 Hours · AI Uplift Person
Make a written guide that takes an extremely smart person (e.g., the median 80k employee) from "I use Claude Chat to assist me with some things" to "man, I can't imagine not trying AI for absolutely everything I do" I suspect there's generally-applicable intuitions on getting…
Francesca Gomez · Wiser Human · AI safety researcher
Evaluate which intervention approaches would get cooperative AI cyber defenses adopted: who is most exposed, which deployment routes are most tractable, which funding models move fastest, and how we'd measure whether it's working.
Mark Aiken · Manager
This project will help develop a coherent framework for recruitment in AI safety, by modelling competency based approaches to defining positions and recruitment processes based on best practices. In one week a participant would ship a usable toolkit: (1) a job analysis and…
Garrett Xu · Brown AI Safety · Executive — Policy Lead
Curriculum and workshops (~2 people, ~2 days): Help design Brown AI Safety's Policy curriculum and workshop series (e.g. memo writing workshop and one-pagers workshop) for Brown AI Safety, with materials other university groups could adapt. Brown AI Safety Practical Policy (~2-3…
Srijit Paul · AI Safety Cardiff University · President
Ship the operating system for a new AI safety hub in Wales: an intake process for its free applied research service, a one-page explainer for prospective partners, and a first draft of the protective assessment framework researchers would use.
Noah Erlwein · Invariant Systems, Inc. · Founder
Build a scenario corpus and minimal evaluator that test whether current agent frameworks catch actions taken outside the intent a user actually granted, and report which categories they miss.
Peter Kazys · SAP Concur · Client Success Manager
Run a controlled, synthetic-data experiment measuring how much modern AI lowers the cost, time and skill needed to re-identify "anonymized" broker records, and turn the number into a threat model and policy brief.
Andrew Nelson · Safe Physical AI (working name) · Cofounder, general leadership and product
Build a comprehensive field map of Physical AI: robotics manufacturers, model developers, evals orgs and safety researchers, so a new initiative can judge neglectedness and target the right partners.
Amir Mohammad Farhang · Compex · Founder & CEO, Compex; organizer, FinBoundBench (ICAIF 2026 competition track)
Build the human-facing operations layer for FinBoundBench, an open benchmark competition launching Sept 1: participant onboarding flow, a support runbook for the six-week development phase, and an outreach tracker for labs, fintech teams, and governance research groups.
Yaseen Ismail · Independent technical research · Ex-Yale researcher
Build a linear, gamified skill-tree style pick-your-path system: an organised route from first hearing about AI safety to the different goals and paths you may reach for, and what you need on the way.
Helen Adepoju · Founder, AI Safety Navigator
Scale a filterable directory of AI safety opportunities so people can find fellowships, training and roles that fit their background, time and timeline, and help find an org to own it long term.