Events, programs, and opportunities from MAIA. Join mailing list
AI Safety Fellowships
Fellowships
Fellowships are the main way people get full-time jobs in AI safety. They pair you with an experienced mentor and fund you to work on a real research project for a few months, and many of the researchers now at organizations like Anthropic, Redwood Research, and Apollo Research got their start through one of the programs below. Beyond the research itself, they are one of the best ways to build a track record, get publications, and meet people in the field. If you are considering a career in AI safety, applying to a fellowship is often the highest-value next step.
CBAI Research Fellowship Featured
In person · Cambridge, MA · ~9 weeks · Fully funded
An intensive, fully-funded AI safety research fellowship run by the Cambridge Boston Alignment Initiative, spanning technical and governance topics like interpretability, multi-agent safety, and formal verification. Open to undergraduates, graduate students, postdocs, and recent graduates, with mentorship from established researchers, a stipend, housing, meals, and compute. Cohorts typically run in the fall.
Fellowship page 
SPAR
Remote · Part time · 3 months
A part-time, fully remote research fellowship run by Kairos that pairs aspiring researchers with experienced mentors on projects in AI safety, security, interpretability, and policy. Open internationally at 5–40 hours per week with no prior research experience required, making it a great option alongside classes. Rounds run twice a year.
Fellowship page 
MATS
In person · Berkeley, CA / London, UK · ~3 months · Funded
An independent research fellowship that pairs researchers with experienced mentors in AI alignment, interpretability, security, and governance. Includes a stipend, housing, compute, and dedicated research management; strong scholars can extend into 6–12 months of funded research. Cohorts run in winter and summer cycles.
Notable alumni include Marius Hobbhahn (co-founder & CEO of Apollo Research) and Thomas Larsen (researcher at the AI Futures Project, coauthor of AI 2027 and lead author of AI 2040: Plan A). Around 10% of MATS alumni have co-founded AI safety organizations.
Fellowship page 
Astra Fellowship
In person · Berkeley, CA · 5 months · Funded
Constellation’s flagship fellowship pairing emerging researchers with senior AI safety mentors at its Berkeley research center, with empirical and strategy/governance tracks. Fully funded with a generous stipend and compute budget; prior AI safety experience is not required. Cohorts run roughly twice a year.
Notable alumni include Aryan Bhatt (Senior Member of Technical Staff at Redwood Research, where he leads high-stakes control research) and Eli Lifland & Romeo Dean (AI Futures Project, coauthors of AI 2027). Over 80% of the first cohort now work full-time in AI safety.
Fellowship page 
Redwood Research Internship
In person · Berkeley, CA · Paid
Redwood Research frequently takes interns (particularly undergrads) to work on AI control, alignment, and futurism research. Being an intern at Redwood Research is broadly similar to participating in other fellowships like MATS or Anthropic fellows, in terms of mentorship, compute and compensation. There is no separate internship application; to intern, just apply through the regular Member of Technical Staff application.
Redwood Research careers 
Anthropic Fellows Program
Full time · 4 months · Funded
A full-time research fellowship in which external engineers and researchers work on Anthropic’s highest-priority AI safety questions, including scalable oversight, AI control, mechanistic interpretability, and model welfare, with close mentorship from Anthropic researchers. Selection emphasizes technical ability over credentials, and many fellows later join Anthropic full-time.
Fellowship page 
OpenAI Safety Fellowship
Remote or Berkeley, CA · ~5 months · Funded
OpenAI’s program for external researchers and engineers to pursue high-impact research on the safety and alignment of advanced AI systems, with priority areas including safety evaluations, robustness, scalable mitigations, and agentic oversight. Fellows work closely with OpenAI mentors and receive a stipend and compute support.
Fellowship page 
Pivotal Research Fellowship
In person · London, UK · 10 weeks · Fully funded
A fully funded research fellowship focused on reducing risks from transformative technology, with tracks in AI safety, AI policy, and biosecurity. Fellows work closely with experienced mentors and receive a stipend plus travel and housing support. Cohorts run twice a year.
Fellowship page 
ERA Fellowship
In person · Cambridge, UK · 8–10 weeks · Fully funded
A fellowship focused on understanding and reducing risks from increasingly capable frontier AI systems, spanning both technical AI safety and governance research. Fellows work closely with mentors; alumni have gone on to roles at the UK AI Security Institute, RAND, Anthropic, and Google DeepMind.
Fellowship page 
LASR Labs
In person · London, UK · 13 weeks · Funded
A full-time technical AI safety research program at the London Initiative for Safe AI (LISA), focused on concrete, action-relevant threat-model research. Participants work in teams of 3–4 under an experienced supervisor, taking a project from proposal to a publishable paper, with a stipend included.
Fellowship page 
ARENA
In person · London, UK · 4–5 weeks · Costs covered
An intensive technical bootcamp at the London Initiative for Safe AI (LISA) that teaches the ML engineering skills needed for technical AI safety research, covering transformers, mechanistic interpretability, and RL. Aimed at people with strong Python and math fundamentals; travel, accommodation, and meals are covered.
Fellowship page