AI Safety Law & Policy Research Fellowship

AI law and policy is a young field with more open questions than people working on them. This fellowship gives a small, selective cohort of Harvard students the time, structure and feedback to take on one of those questions and answer it well.

Each fellow writes one original paper of 6,000–12,000 words. Student research managers keep each project on track, expert reviewers read the drafts, and fellows meet every week to write together. The program is run jointly by the AI Safety Team (AIST) and the Harvard Law AI Student Association (AIA).

Fellows leave with a polished writing sample suitable for job applications, a clearer sense of whether a career in AI law and policy suits them, and a working understanding of why AI safety matters.

Over twelve weeks, each fellow submits four deliverables.

• Thesis statementWeek 2The argument you will defend, in a paragraph.
• Full outlineWeek 4Structure, sources and the strongest objections.
• Rough draftWeek 8An expert reviewer reads it and sends written feedback.
• Final draftWeek 12Reviewed again and considered for the best-paper prize.

Alongside the writing

• Weekly writing group meetings at the AIST offices, with food
• A policy research seminar
• A closing event for all fellows, where the best-paper prize is presented

The questions below are examples only. The list is not exhaustive, and we encourage applicants to propose their own topics.

1. Should DOJ issue a waiver, or Congress enact an exemption, protecting AI labs that jointly negotiate minimum safety standards from antitrust liability?
2. How should Congress design, structure and delegate to a federal AI regulatory agency?
3. Should frontier AI companies be legally required to report users who show signs of planning mass-casualty attacks, and on what terms?
4. What safe harbor should protect independent AI safety testers from CFAA, contract and tort liability?
5. Can the government regulate AI “character” traits without violating the First Amendment?
6. How should an AI agent determine which jurisdiction’s law governs its actions, and who is liable when it breaks that law?
7. Can securities disclosure rules force meaningful transparency about catastrophic-risk practices at publicly traded AI labs?
8. Should Congress codify DoD Directive 3000.09’s “appropriate human judgment” standard, and what should that statutory standard require?

The fellowship is open to all Harvard affiliates. It is designed primarily for graduate students, with one or two places reserved for exceptional undergraduates. The cohort will number six to twelve fellows, and admission is highly selective.

Each fellow receives a stipend of $4,000, paid in two installments:

• At the start of the fellowship$1,000
• On completion of the final paper$3,000

A prize is awarded for the best paper, and selected papers will be put forward for publication with outside outlets.¹

1. Submit the written application, including a writing sample.
2. Shortlisted candidates are invited to interview.

Trusted individuals and partner organizations may also refer applicants.