OpenAI’s first-ever Safety Fellowship has already closed its application window; the deadline was 3rd May 2026, but if you applied, you’re not far from an answer: OpenAI has committed to notifying successful applicants by 25th July 2026, just a few weeks from now.
This is a genuinely unusual opportunity in the AI research funding landscape, not because of the money involved. However, that’s substantial, but because of who it’s actually built for: people without a PhD, without published papers, and without prior machine learning experience are explicitly told they can apply. If this is your first time hearing about it, this guide covers everything the fellowship involves, so you’re ready if and when OpenAI opens a future round.
The OpenAI Safety Fellowship is a pilot program created to support the next generation of AI safety researchers. Fellows will work closely with OpenAI mentors and a cohort of other participants. The programme focuses on important safety questions for current and future AI systems.
OpenAI Safety Fellowship Summary
| Fellowship Name ⇒ | OpenAI Safety |
| Host Country ⇒ | United States |
| Study Level ⇒ | Fellowships |
| Benefits ⇒ | $3,850 weekly stipend + up to $15,000/month compute support + mentorship + workspace access + possible visa support |
| Funded by ⇒ | OpenAI |
| Eligible Countries ⇒ | All countries |
| Application Deadline ⇒ | Closed 3rd May 2026. Notifications due 25th July 2026. This is a pilot cohort — no confirmed date for a future round. |
About the OpenAI Safety Fellowship
This is a pilot programme, meaning it’s the first time OpenAI has run something in exactly this form, aimed at bringing external researchers, engineers, and practitioners into rigorous AI safety and alignment work without requiring them to already work inside a major AI lab. It runs for roughly five months, from 14th September 2026 to 5th February 2027, based primarily out of Constellation, a shared research workspace in Berkeley, California, though remote participation is explicitly permitted for anyone who can’t relocate.
What sets this apart from a typical academic fellowship is how OpenAI frames its selection priorities: research ability, technical judgment, and execution matter more than formal credentials, and the programme is deliberately open to people from computer science, cybersecurity, social science, privacy, human-computer interaction, mathematics, and physics backgrounds, not just machine learning specialists. Fellows work with open-source models and public APIs rather than OpenAI’s internal systems, and each fellow is expected to produce a substantial, concrete research output by the end of a paper, a benchmark, a dataset, or something comparable rather than simply contributing to an existing project.
What the Fellowship Provides
- A weekly stipend of $3,850.
- Compute support of up to approximately $15,000 per month.
- Direct mentorship from members of OpenAI’s safety and alignment teams.
- Workspace access at Constellation in Berkeley, California, for fellows who choose to work in person.
- A remote participation option for those who can’t relocate.
- Possible visa support for selected fellows, including J-1, F-1 CPT (full-time only), and F-1 OPT with Designated School Official approval, among other arrangements considered case by case.
- API credits and additional research resources, though fellows won’t have access to OpenAI’s internal systems.
- Access to a cohort of peers and a genuine mentorship network within the AI safety research community.
Eligibility Requirements
- You must be genuinely interested in full-time, empirical AI safety and alignment research.
- You need to be available full-time for the entire fellowship period, roughly 40 hours a week from September 2026 to February 2027.
- You need strong technical skills, particularly in Python, and the ability to work through difficult, poorly defined technical problems.
- You should be capable of working independently while also collaborating well within a cohort.
- You need a genuine interest in at least one of the fellowship’s priority research areas.
- You must be willing and able to produce a substantial final research output by the programme’s end.
- You need at least two references who can speak credibly to your strengths and experience.
- You do not need a PhD, prior machine learning work experience, or published academic papers — OpenAI is explicit that these aren’t prerequisites.
Priority Research Areas
- Safety evaluation
- AI ethics
- Robustness and reliability
- Scalable safety mitigations
- Privacy-preserving safety methods
- Agentic oversight
- High-severity misuse prevention
- AI alignment research
One important note: OpenAI’s application system doesn’t allow edits after submission, so save a copy of your responses before you submit, and review everything carefully first.
Building Your Application File
1. Completed online application form.
This is submitted through OpenAI’s official fellowship portal, and it’s worth knowing upfront that once you hit submit, your answers are locked — there’s no editing afterwards. Draft your responses somewhere else first, review them carefully, and only paste them into the live form when you’re genuinely satisfied, rather than composing directly in the application window.
2. Detailed, specific answers to the application’s substantive questions.
This is where most applications actually get decided, and it’s also where generic responses do the most damage. OpenAI is evaluating research ability and technical judgment, not enthusiasm alone, so an answer like “I’m passionate about AI safety and want to make AI beneficial for humanity” tells the reviewer nothing they can act on.
3. At least two references who know your work well.
These need to be people who’ve directly observed you solving hard problems, not just people with impressive titles who know your name. A research supervisor, a manager who watched you debug a genuinely difficult issue, or a close collaborator on a technical project will write a far more useful reference than a senior figure who only knows you in passing.
4. Information about your technical background and research interests.
This section is where you connect your actual skills — Python proficiency, any experience with the priority research areas, relevant coursework or self-directed study — to the kind of work the fellowship expects you to do. Be concrete about what you’ve built or worked on rather than listing skills abstractly; “implemented and evaluated three jailbreak-detection methods on an open-source model” tells a reviewer far more than “experienced in AI safety.”
5. Details of your preferred research area.
Choose from the fellowship’s priority areas — safety evaluation, AI ethics, robustness and reliability, scalable safety mitigations, privacy-preserving safety methods, agentic oversight, high-severity misuse prevention, or AI alignment research — and explain specifically why that area fits your background and interests, rather than picking the one that sounds most impressive.
6. Visa information, if applicable to your situation.
If you’re not a U.S. citizen or permanent resident, be ready to specify your current visa status and, where relevant, which of OpenAI’s supported visa routes might apply to you — J-1, F-1 CPT (full-time only), or F-1 OPT with Designated School Official approval.
The Selection Process
Selection runs in three stages: first, a review of your application and references; second, technical assessments and interviews for those who pass the initial screen; and third, a final interview with a potential OpenAI mentor. This is a genuinely multi-layered process, so candidates who make it past the first stage should expect meaningful additional time investment before a final decision.
How to Prepare Before the Next Application Cycle
Since this is a pilot programme, there’s no confirmed date for round two, but that actually works in your favour if you start now, since most applicants scramble to pull a strong application together in the final weeks of a window. Here’s where to focus in the meantime:
1. Pick a research direction and start building evidence of real work in it.
OpenAI weighs technical judgment and execution over credentials, so the strongest thing you can do isn’t wait for the fellowship to start thinking about AI safety — it’s picking one of the priority areas (safety evaluation, robustness, agentic oversight, scalable mitigations, and so on) and producing something concrete now: a small experiment, a write-up, a piece of open-source analysis.
2. Sharpen your Python and empirical research skills specifically.
The fellowship is explicit about wanting technically strong, empirically grounded work. If your background is more theoretical or you’re coming from a non-CS field like physics or social science, use this time to get comfortable running real experiments and working with public model APIs, since that’s the actual day-to-day of the role.
3. Read OpenAI’s and other labs’ published safety research closely.
Understanding what “safety evaluation” or “scalable safety mitigations” actually mean in current practice, not just as buzzwords, will make your eventual application answers sharper and more specific, and it’ll help you identify which research area genuinely fits your interests rather than picking one blindly.
4. Start cultivating your two references now, not the week you apply.
OpenAI wants references who can speak credibly to your strengths and experience. That means people who’ve actually seen you work, a professor who supervised a project, a manager who watched you solve hard problems, a collaborator on open-source work. Strong references take relationship-building time, not a last-minute request.
5. Draft your research interest statement early and revise it over time.
Since your application answers can’t be edited after submission, treat your first draft as a working document, not a final one. Write out your research interests and technical background now, then revisit and tighten it every few weeks as your thinking develops.
6. Follow OpenAI’s alignment and safety channels for the actual announcement.
Since there’s no fixed calendar here, the only reliable way to catch the next window in time is to watch OpenAI’s official alignment blog and safety announcements directly, rather than waiting to hear about it secondhand once the deadline is already close.
7. Build a portfolio piece you can point to.
A benchmark you built, a small dataset you curated, an analysis you published informally, having one tangible thing to reference in your application does more to demonstrate execution ability than a paragraph describing your potential.
Frequently Asked Questions
Is the OpenAI Safety Fellowship still accepting applications?
No. The deadline was 3rd May 2026, and applicants are waiting on notifications due by 25th July 2026. Since this is a pilot programme, there’s no confirmed date for a future round yet.
Can Nigerian or other African applicants apply for the OpenAI Safety Fellowship?
Yes. The fellowship is open to applicants from any country, and remote participation is explicitly permitted for anyone who can’t relocate to Berkeley.
Do I need a PhD or prior machine learning experience to apply?
No. OpenAI explicitly states that a PhD, prior machine learning work experience, and published academic papers are not required. The programme prioritises research ability, technical judgment, and execution over formal credentials.
Will fellows have access to OpenAI’s internal AI systems?
No. Fellows work with open-source models and public APIs. Internal OpenAI systems remain off-limits throughout the fellowship.
Is remote participation genuinely an option, or is this mainly an in-person programme?
Remote participation is a real option, not a token accommodation. While the programme is primarily based in Berkeley through the Constellation workspace, OpenAI explicitly allows fellows who can’t relocate to participate remotely for the full fellowship period.
Disclaimer: All fellowship details, deadlines, and eligibility criteria in this post were verified from official sources as of July 2026. Information is subject to change without notice. Always confirm the latest details directly on OpenAI’s official Safety Fellowship page before applying. ScholarWaka is not affiliated with OpenAI.
Official Application Portal
Join ScholarWaka Community
Get the latest scholarship updates and study abroad opportunities before anyone else.
