The OpenAI Safety Fellowship 2026 in Berkeley, California is a paid six-month research program for external researchers who want to work on high-impact AI safety and alignment problems. The fellowship runs from September 14, 2026, to February 5, 2027, offers a $3,850 weekly stipend, provides up to about $15,000 per month in compute support, and gives fellows access to workspace in Berkeley or the option to participate remotely.
For people searching terms like OpenAI Safety Fellowship 2026, paid AI safety fellowship in Berkeley, and AI alignment fellowship California, this is one of the most exciting opportunities of the year because it combines direct mentorship, strong financial support, and a real chance to produce a serious research output such as a paper, benchmark, or dataset.
Basic Information about the Fellowship
- Program name: OpenAI Safety Fellowship 2026.
- Host organization: OpenAI.
- Location: Berkeley, California, USA, with remote participation also allowed.
- Program duration: September 14, 2026, to February 5, 2027.
- Duration: September 14, 2026 – February 5, 2027
- Time commitment: Full-time (40 hours/week)
- Funding:
- $3,850 per week stipend.
- Up to approximately $15,000 per month in compute support as needed.
- Deadline: May 3, 2026 (11:59 PM, Anywhere on Earth)
Institution Details
The fellowship is offered by OpenAI, one of the world’s most influential AI research and deployment organizations, with work spanning language models, reasoning systems, safety research, and AI infrastructure. Although the fellowship is associated with Berkeley, California, fellows are expected to work as external researchers rather than enrolled university students, and the program includes access to workspace at Constellation in Berkeley, a nonprofit environment that supports AI safety research.
This is not a university degree scholarship, which makes it different from traditional academic fellowships. Instead, it sits at the intersection of independent research, industry-grade mentorship, and practical AI safety work, making it highly relevant for applicants interested in the future of safe advanced AI systems.
What Fellows Will Do
- Work on high-impact research projects
- Collaborate with OpenAI mentors & peer cohort
- Produce a final output (paper, dataset, or benchmark)
- Use external tools (no internal system access)
Study Areas
This fellowship is not built around traditional academic degree programs. Instead, it supports focused work in AI safety and adjacent research areas. Public sources identify the following key research domains:
- AI safety & alignment research
- Safety evaluation
- Ethics
- Robustness
- Scalable mitigations
- Privacy-preserving methods
- Agent oversight
- Preventing high-risk misuse
This wide scope is one reason the fellowship is appealing. It is not only for traditional machine learning researchers; OpenAI also appears open to applicants from computer science, cybersecurity, HCI, and social science backgrounds, as long as they can contribute meaningfully to real AI safety problems.
OpenAI Safety Fellowship Benefits
- Stipend: $3,850 per week
- Compute support: up to ~$15,000/month
- Mentorship from OpenAI researchers and staff.
- Workspace in Berkeley (optional remote work)
- Remote participation option, which increases accessibility for global applicants.
- Visa support available (J-1, F-1/CPT, OPT, etc.)
- API credits and research resources
- Peer cohort environment with other fellows.
- Opportunity to produce a significant research output, such as a paper, benchmark, or dataset.
- Direct exposure to frontier AI safety questions that matter for real-world systems.
This is one of the biggest strengths of the program: it is not only paid, but also designed to help fellows do meaningful work instead of just observing from the sidelines.
Benefits not included
- This is not a university degree scholarship, so it does not include tuition funding for a bachelor’s, master’s, or PhD.
- Fellows do not get access to OpenAI’s internal systems, according to one report summarizing the program.
- Housing is not clearly confirmed as fully covered in the reviewed sources, even though Berkeley workspace is provided.
- Family or dependent support is not mentioned in the available program summaries.
- Visa and relocation details are not clearly described in the reviewed public summaries, so applicants should not assume automatic sponsorship unless confirmed during selection.
- Private lifestyle or personal travel expenses are not listed as covered.
Eligibility Criteria for the OpenAI Safety Fellowship
The reviewed sources suggest that OpenAI is prioritizing ability and execution over formal academic pedigree:
- Backgrounds: CS, math, physics, cybersecurity, social science, etc.
- No PhD or publications required
- Strong Python coding skills
- Ability to solve ambiguous technical problems
- Interest in empirical AI safety research
- Good collaboration & execution skills
- Applicants should demonstrate strong research ability, technical judgment, and execution capacity.
- Relevant backgrounds can include computer science, cybersecurity, social science, and human-computer interaction.
- Fellows must be able to produce a substantial output by the end of the fellowship.
Interview Process
- Application + reference review
- Technical assessment & interviews
- Final live interview with mentor(s)
Key Notes
- Must submit specific, concrete answers
- Cannot edit application after submission
- AI-generated answers may appear too generic (use carefully)
Selection Criteria
- Research ability
- Technical judgment
- Execution skills
- Not focused on credentials or writing quality
Required Documents for the Application
The public summaries do not show a full official checklist in one place, but the reviewed sources make it clear that applicants should be ready to submit materials that demonstrate strong execution and research fit:
- Completed application through the OpenAI fellowship process.
- Personal and professional background information.
- Evidence of technical or research ability.
- Project or research alignment details, showing how the applicant fits the fellowship’s AI safety priorities.
- Supporting documents or work samples strong enough to demonstrate execution and research quality.
- Any requested application documentation submitted by the deadline of May 3, 2026.
Since OpenAI emphasizes output and execution, applicants should assume that a strong portfolio, serious work samples, and clearly framed research ideas will matter more than generic academic paperwork.
OpenAI Safety Fellowship Application Process
- Step 1: Review the OpenAI Safety Fellowship announcement and understand the program timeline, goals, and research areas.
- Step 2: Decide whether your background fits AI safety, alignment, robustness, privacy, governance, cybersecurity, HCI, or related domains.
- Step 3: Prepare your application materials with a strong focus on research ability, technical execution, and project fit.
- Step 4: Submit your application before May 3, 2026.
- Step 5: Wait for the review process while OpenAI assesses applicants based on their ability to work on real-world AI safety questions.
- Step 6: Successful applicants are expected to be notified by July 25, 2026.
- Step 7: If selected, prepare for either in-person work in Berkeley or remote participation, depending on program arrangements.
- Step 8: Begin the fellowship on September 14, 2026, and work toward a substantial final output by February 5, 2027.
Tips for a Successful Application
- Focus on execution, not just credentials, because OpenAI’s public messaging clearly emphasizes capability over prestige.
- Show real work, such as research papers, benchmarks, experiments, security work, or thoughtful safety analysis.
- Target a concrete AI safety problem, rather than presenting a broad interest in AI ethics.
- Demonstrate technical judgment, especially if your work touches robustness, misuse, privacy, or oversight.
- Make your application output-oriented, since fellows are expected to deliver a paper, dataset, or benchmark.
- Apply early, because the deadline is tight and strong applications usually need revision.
- If you come from a nontraditional background, explain your edge clearly, because the program appears open to diverse disciplines beyond standard ML research.
Monthly Living Expenses for International Students in Berkeley, California
Although this is a fellowship rather than a university scholarship, applicants still need a realistic Berkeley budget:
- Accommodation: USD 1,400 to USD 2,600 per month depending on shared housing or private rental.
- Food and groceries: USD 350 to USD 650 per month.
- Transportation: USD 80 to USD 180 per month.
- Utilities and internet: USD 120 to USD 220 per month.
- Phone and personal expenses: USD 120 to USD 300 per month.
- Estimated total monthly cost: USD 2,070 to USD 3,950 in Berkeley depending on lifestyle.
With a stipend of $3,850 per week, the fellowship funding is very strong relative to normal living costs, especially for participants who use shared housing and budget wisely.
Why Choose This Fellowship?
- It is paid generously, with a $3,850 weekly stipend and additional compute support.
- It focuses on one of the most important topics in modern technology: AI safety.
- It is offered by OpenAI, which gives fellows direct exposure to frontier research thinking.
- It values skill and execution over formal status, making it more accessible to talented nontraditional applicants.
- It supports serious research outputs, not just networking or passive learning.
- It allows Berkeley-based or remote participation, which adds flexibility.
- It can become a major career signal for people working in AI alignment, safety, policy, or technical governance.
Frequently Asked Questions about the fellowship
- Is the OpenAI Safety Fellowship 2026 paid?
Yes, fellows receive $3,850 per week, along with project-related compute support that can reach about $15,000 per month as needed. - When does the fellowship run?
The program runs from September 14, 2026, to February 5, 2027. - What is the application deadline?
The application deadline is May 3, 2026. - Do I need a PhD to apply?
No specific PhD requirement is emphasized in the reviewed sources, and OpenAI appears to prioritize ability, judgment, and execution over formal credentials. - Is the fellowship only for computer scientists?
No, related fields like cybersecurity, social science, and human-computer interaction are also mentioned as relevant backgrounds. - Can I join remotely?
Yes, public reports state that remote participation is allowed, although workspace in Berkeley is also available. - What do fellows have to produce?
Fellows are expected to produce a substantial output such as a paper, benchmark, or dataset by the end of the fellowship.
