How to get into AI safety

There is no single entry point, but there is a common order: take a course first, then a part-time project, then a full-time fellowship. Every step lists only high-quality programs.

Courses and technical training

Start here: exercises or discussion rather than survey material.

Frontier AI Safety Course (Chinese)Our own courseThe Chinese edition of the BlueDot Impact curriculum: six weeks of seminar-style co-learning across AGI strategy, technical safety and frontier governance.

BlueDot Impact · Technical AI SafetyMulti-week courseThe technical track; the same platform runs AGI Strategy and governance courses.

BlueDot Impact · Frontier AI GovernanceGovernance courseFree; a six-day intensive or a six-week part-time cohort, ending in a concrete plan for your own contribution.

ARENA CurriculumTechnical trainingTransformer interpretability, reinforcement learning and LLM fundamentals with exercises.

Part-time and remote

Then a small project: lighter weekly load, remote-friendly.

SPARRemote research programPart-time and remote; supervised alignment research for technical people moving into research.

Full-time fellowships

Then make it a career: around 40 hours a week, most require being on site.

Anthropic Fellows ProgramPaid fellowship4 months of empirical safety research inside Anthropic, recruited in cohorts.

MATS ResearchResearch fellowship12 weeks at 40 hours a week, on site in Berkeley or London, with a mentor.

Cambridge ERA:AI FellowshipResearch fellowship10 weeks in Cambridge with a £10,000 stipend and travel covered; 200+ researchers supported so far.

Astra FellowshipFunded fellowship3–6 months for people researching AI safety full-time, with funding and a peer network.

Singapore AI Safety FellowshipResidential fellowship3 months in Singapore with SGD 5,000 a month plus housing; first cohort closed, next in spring/summer 2027.

GovAI · Centre for the Governance of AIGovernance fellowship3 months on site in London, research and applied tracks; two cohorts a year, the 2027 winter cohort runs 18 January – 9 April.

Inclusion: technical (alignment research, interpretability, safety engineering, evaluations) or governance (research, standards, international cooperation), and demonstrably high quality; nothing we cannot verify.