The Federation of American Scientists (FAS) is accepting applications for a special AI Safety cohort of its Policy Entrepreneurship Fellowship, a part-time programme running from 30 September 2026 to 28 February 2027. The fellowship is designed for people who already understand a meaningful frontier-AI safety, security or governance problem and want to move beyond analysis toward a policy solution that could realistically be adopted.
Selected fellows receive a $5,000 stipend, structured policy-entrepreneurship training, individual mentorship, professional editing support and access to FAS experts and networks. One fellow may also receive an additional $1,000 merit award for the most specific and actionable policy memo and advocacy plan. Applications close at 11:59 p.m. Eastern Time on 7 September 2026.
Fellowship at a Glance
- Organization: Federation of American Scientists (FAS)
- Programme: AI Safety Policy Entrepreneurship Fellowship
- Format: Part-time, hybrid fellowship
- Dates: 30 September 2026 – 28 February 2027
- Average commitment: 5 hours per week / 20 hours per month
- Stipend: $5,000
- Additional award: Up to $1,000 merit award for one fellow
- Travel support: Additional stipend for selected retreat and capstone expenses
- Deadline: 7 September 2026, 11:59 p.m. ET
What Makes This Fellowship Different
This is not primarily a research fellowship in which participants spend several months writing a paper and then stop. FAS explicitly frames the programme around policy entrepreneurship: identifying a real policy window, understanding who has the authority to act, building coalitions, engaging decision-makers and moving a proposal from an idea toward implementation.
That distinction matters. A successful applicant does not need to arrive with a complete legislative package, but they should be able to identify a concrete problem in frontier AI and explain what type of intervention could plausibly address it. The programme then helps fellows turn that intervention into a policy memo and an advocacy or implementation strategy.
Frontier AI Topics FAS Wants Fellows to Work On
The cohort is focused on frontier AI safety and the governance of advanced AI systems. FAS lists several areas of interest, including technical AI safety, risk assessment, transparency, accountability and policy frameworks for safe and secure AI development and deployment.
More specific themes include:
- Cybersecurity risks connected to advanced AI systems.
- Capability thresholds and how governments or institutions should respond when models cross them.
- Risk taxonomies for frontier models.
- AI risks to critical infrastructure.
- Compute governance and oversight of advanced computing resources.
- Model proliferation and policy questions around open versus closed systems.
- Institutional capacity for governing catastrophic or systemic AI risks.
- Standards, procurement rules and regulatory mechanisms for high-risk AI systems.
Applicants are not restricted to these examples, but the proposal should clearly connect to frontier AI safety, security or governance and should have a realistic path to policy adoption.
The Core Output: A Policy Memo That Can Move
Each fellow will develop and publish a policy memo focused on a specific frontier-AI safety challenge. The memo needs to do more than explain why the problem matters. It should recommend a practical solution and identify the institutions, decision-makers and stakeholders that would need to act.
The adoption route could involve legislation, regulation, standards, government procurement, institutional design or another governance mechanism. FAS is interested in proposals that show an understanding of implementation rather than treating policy as an abstract recommendation.
Training and Reading Groups
Fellows participate in policy-entrepreneurship training sessions and reading groups. These sessions are intended to deepen understanding of both frontier AI safety and the mechanics of policymaking: how proposals gain attention, how coalitions are built, how technical expertise is translated for policy audiences and how an idea can be positioned for adoption.
There are also twice-monthly group check-ins. FAS says it will make an effort to schedule those meetings at times that accommodate participants, with limited absences considered when appropriate notice is given.
California Retreat and Washington DC Capstone
The fellowship contains two important in-person components. A California retreat is tentatively scheduled for 4–7 November 2026. It will focus on mentorship, community building and advancing each fellow’s policy project.
A capstone event in Washington, DC is tentatively planned for the week of 22 February 2027. Fellows will use that event to launch and disseminate their policy memos and interact with policymakers, experts and other relevant stakeholders.
FAS provides an additional travel stipend to cover selected expenses connected with the retreat and capstone. Applicants should still read the final participation instructions carefully before accepting an offer, particularly if international travel or visa arrangements would be required.
Who Is the Fellowship Designed For?
FAS encourages applicants from different professional backgrounds. The programme is particularly suitable for early- to mid-career professionals who have technical or subject-matter expertise but relatively limited direct experience working in public policy.
Potential backgrounds include:
- Technical AI research and machine learning.
- Academic research.
- Think tanks and civil society.
- Technology industry roles.
- Law and regulation.
- Cybersecurity.
- National security.
- Science and technology policy.
- Other disciplines with a credible connection to frontier AI risk.
This makes the fellowship especially relevant to someone who can already explain an AI safety problem in depth but wants to learn how to influence government or institutional action.
What FAS Will Look for in Applications
Applications are evaluated holistically. FAS specifically wants evidence that the applicant understands a significant frontier AI safety, security or governance challenge and has thought seriously about how a policy response could work in practice.
Strong applications should demonstrate:
- A clearly defined AI safety or governance problem.
- A concrete policy idea rather than a broad call for “more regulation”.
- A credible route toward real-world adoption.
- Awareness of the institutions with authority over the issue.
- Thoughtful identification of decision-makers and stakeholders.
- Understanding of practical barriers to policy change.
- Initiative, curiosity and willingness to engage constructively with those barriers.
How to Choose a Strong Policy Idea
A useful proposal is narrow enough to act on. For example, “make AI safer” is too broad. A stronger proposal might focus on a particular reporting standard for frontier-model incidents, a procurement requirement for government use of advanced AI, a mechanism for monitoring high-end compute, a defined approach to critical-infrastructure risk or a standard for evaluating a specific dangerous capability.
The applicant should be able to answer four questions: What exactly is the problem? Who has authority to change the situation? What policy tool should they use? What would make adoption politically or institutionally feasible?
Time Commitment and Mandatory Participation
Fellows must commit an average of five hours per week, approximately 20 hours per month, from 30 September through 28 February. Mandatory FAS programming is part of the fellowship, although exceptional circumstances can be considered for some sessions.
Because this is a part-time programme, it may be compatible with a job, postgraduate study or another professional role, but applicants should account for the in-person events and recurring group sessions before applying.
What Fellows Receive
- $5,000 fellowship stipend.
- Potential $1,000 merit award for the strongest actionable memo and advocacy plan.
- Travel support for selected retreat and capstone expenses.
- Professional editing of fellowship writing.
- Policy-engagement training from FAS staff.
- Monthly one-to-one check-ins with an assigned FAS staff mentor.
- Connections to relevant experts and policy actors.
- Invitations to FAS programming and science-and-technology policy briefings.
How to Prepare Before Applying
Applicants should spend time sharpening the policy concept before opening the application. Write a short internal brief for yourself covering the problem, evidence that it matters, the proposed intervention, the agency or institution that could implement it and the stakeholders that could support or oppose the idea.
If your background is highly technical, translate the risk into policy language without removing the technical substance. If your background is policy or law, demonstrate that you understand the relevant frontier-AI capabilities and risks rather than relying on general AI terminology.
How to Apply
Applications must be submitted through the application platform linked from the official FAS fellowship page. Applicants need to create an account on that platform and complete the fellowship application there. FAS states that it will only accept applications submitted through the linked system.
The deadline is 7 September 2026 at 11:59 p.m. Eastern Time. Questions about the fellowship can be directed to the programme contact listed by FAS.
Frequently Asked Questions
Is the FAS AI Safety Policy Entrepreneurship Fellowship paid?
Yes. Selected fellows receive a $5,000 stipend. One fellow may also receive an additional $1,000 merit award.
How many hours per week does the fellowship require?
FAS requires an average commitment of about five hours per week, or 20 hours per month, across the fellowship period.
Do applicants need previous government policy experience?
No. The fellowship is particularly suitable for early- to mid-career professionals with limited direct policy experience who want to translate technical or subject expertise into policy impact.
Is the fellowship fully remote?
No. The programme is hybrid and includes tentative in-person events in California and Washington, DC, in addition to ongoing fellowship activities.
