MATS Research Fellowship: A Deep Dive into AI Safety and Security
The field of artificial intelligence is advancing at an unprecedented pace, bringing with it both incredible potential and significant challenges. As AI systems become more powerful, ensuring their safety, transparency, and alignment with human values is paramount. The MATS Research fellowship is a leading program designed to cultivate the next generation of experts dedicated to this critical mission. This intensive, fully funded program offers a unique opportunity for aspiring researchers to contribute to AI alignment and security.
Program Overview and Goals
The MATS Research fellowship is a highly competitive, 10 to 12-week in-person program focused on AI alignment, AI security, and reducing risks from advanced AI. It brings together talented fellows and world-leading mentors from top AI institutions. The program aims to provide an immersive research environment, removing financial and logistical barriers so participants can focus entirely on their work. The ultimate goal is to prepare individuals to make meaningful contributions to ensuring that advanced AI systems are safe and beneficial for humanity.
Key Program Details
This fellowship is hosted in major research hubs, offering participants a choice between two prestigious locations: Berkeley, USA, or London, UK. The program is scheduled to begin on September 28, 2026, with an application deadline of June 7, 2026. The MATS Research fellowship provides a comprehensive funding package designed to support fellows fully during their tenure.
Funding Package Inclusions
- Stipend: Fellows receive a $12,500 stipend to cover personal living expenses.
- Compute Budget: A generous $20,000 compute budget is provided, granting access to substantial computational resources essential for AI research.
- Housing: Free housing is provided, eliminating a major financial concern for participants.
- Meals: Meals are covered, ensuring fellows are well-nourished throughout the program.
- Travel Costs: Travel expenses to and from the fellowship location are reimbursed.
- Visa Support: Assistance with obtaining a J-1 visa is offered for international participants, if required.
Research Tracks Available
The fellowship offers a diverse range of research tracks, allowing fellows to align their work with their specific interests and skills. Participants can apply to one or multiple tracks, depending on their background and aspirations.
Areas of Focus
- Empirical Research: This track focuses on conducting experiments and analyzing data to understand AI behavior and safety properties.
- Theoretical Research: Fellows in this track explore the fundamental principles and mathematical underpinnings of AI alignment and safety.
- Strategy and Forecasting: This area involves developing strategies and making predictions about the future development and impact of AI.
- Policy and Governance: Participants examine the legal, ethical, and societal frameworks needed to guide AI development and deployment.
- Systems Security: This track concentrates on securing AI systems against malicious attacks and ensuring their robustness.
- Field-building and Founding: This track supports individuals interested in establishing new organizations or initiatives within the AI safety community.
- Biosecurity: This emerging area explores the intersection of AI and biological safety, including potential risks and mitigation strategies.
Each track is carefully designed to address different facets of AI safety, from highly technical research to broader analyses of societal impact and governance.
Mentorship and Research Environment
A cornerstone of the MATS Research fellowship is its robust mentorship model. Fellows are paired with experienced researchers from leading AI labs and safety organizations, including Anthropic, OpenAI, Google DeepMind, Redwood Research, and ARC. These mentors provide invaluable guidance throughout the program.
Mentor Support
- Direct Research Guidance: Mentors offer hands-on advice and direction for research projects.
- Project Direction: They help shape research questions, methodologies, and project scope.
- Technical and Conceptual Support: Mentors provide expertise on complex technical challenges and theoretical concepts.
- Publication and Scaling: Support is offered for developing research ideas into publishable papers and scaling their impact.
This structured mentorship ensures fellows can rapidly advance their research skills while contributing to meaningful AI safety work.
Fellowship Benefits
The MATS Research fellowship is designed to provide a comprehensive support system, allowing fellows to dedicate their full attention to research and professional development.
Financial and Material Support
Beyond the stipend and compute budget, fellows benefit from free accommodation and meals, as well as reimbursement for travel costs. This holistic financial support system removes many common barriers faced by researchers.
Academic and Professional Growth
Participants gain direct collaboration opportunities with top AI researchers, access to cutting-edge problems in AI safety, and the chance to publish their findings. The program also offers significant exposure to frontier AI labs and institutions, which can open doors to future career opportunities.
Community and Networking
The fellowship fosters a strong sense of community through its cohort-based learning environment. Fellows connect with a global network of AI safety researchers and professionals, building relationships that can last throughout their careers. This network is invaluable for collaboration and knowledge sharing within the AI governance and technical alignment fields.
Application Process
The application process for the MATS Research fellowship is multi-stage, designed to identify candidates with strong technical abilities and significant research potential.
Stage 1: General Application
This initial stage is open until June 7, 2026, and typically takes 1-2 hours to complete. It involves answering background questions and providing track-specific responses. Applicants will also need to supply contact details for two references.
Stage 2: Central Evaluation
Candidates who successfully pass Stage 1 proceed to standardized assessments. The nature of these assessments varies by track. For example, the Empirical Track includes tests for research taste and technical implementation, while the Policy & Governance Track requires an analytical writing evaluation. Other tracks may move directly to Stage 3.
Stage 3: Stream Applications
In this stage, applicants rank their preferred research streams and submit tailored responses or research proposals. This may also involve interviews or specific work tests designed to gauge suitability for particular research areas.
Final Offers
Decisions are typically finalized in late July or early August 2026. The matching process carefully considers both mentor and applicant preferences to ensure a good fit for all parties involved. The program officially begins on September 28, 2026.
Impact and Outcomes
The MATS Research fellowship has a proven track record of success within the AI safety ecosystem. A significant majority of its alumni continue to work in fields related to AI alignment, transparency, or security. The program has contributed to over 190 research publications in recent years, accumulating more than 12,000 citations. Graduates are well-represented in leading AI organizations and academic institutions, demonstrating the program’s effectiveness in nurturing talent.
Why This Fellowship Matters
As AI systems become increasingly sophisticated and capable, the need for rigorous safety research intensifies. The MATS Research fellowship directly addresses this need by developing skilled researchers in AI alignment, supporting early-career talent in high-impact research, and strengthening the global AI safety infrastructure. It serves as a vital bridge connecting academia, industry, and policy communities, fostering a collaborative approach to ensuring AI’s beneficial future.
Frequently Asked Questions
What is the MATS Research Fellowship?
The MATS Research Fellowship is a 10-12 week program designed to train researchers in AI safety, alignment, and security, offering full funding and mentorship.
Where is the fellowship located and when does it start?
The fellowship is hosted in Berkeley, USA, or London, UK, and begins on September 28, 2026.
What kind of financial support does the fellowship provide?
It includes a $12,500 stipend, a $20,000 compute budget, free housing, covered meals, and reimbursed travel costs.
What are the different research tracks available?
Available tracks include Empirical Research, Theoretical Research, Strategy and Forecasting, Policy and Governance, Systems Security, Field-building and Founding, and Biosecurity.
