Help support the next generation of AI safety talent

AI Safety Global Society is a nonprofit organisation helping people transition into high-impact research careers. We run upskilling programs that train and mentor early to mid-career researchers working to reduce existential risks from transformative AI. 

We’re looking for experienced people to support this work as teaching assistants, research leads, and facilitators. We are currently applying for funding. This is an expression of interest for anyone who would like to be involved in the first iteration of our programs.

Open Roles

Research Leads

Program: AI Control Research Fellowship

Duration: 8 weeks
Approx dates: TBC
Time commitment: ~10 hours/week
Format: Remote, part-time

You will support early to mid-career researchers in researching AI control topics, such as monitoring, control evaluations, or capability control.

Responsibilities:

• Lead a small research group (3–5 fellows)
• Help design and scope projects
• Coordinate with the AISGS team and fellow mentors
• Review and guide research outputs
   

Ideal background:

• Ability to understand and explain at least 30% of the research mentioned Ram Potham’s literature review or Ryan Greenblatt's overview of areas of Control work
• Technical research experience (does not need to be in AI control specifically)
• Ability to support independent research and experimental design

• Comfortable giving high-level guidance on technical safety questions

   

Facilitators

Program: AI Control Reading Group

Based on: Louis Thomson Syllabus, with reference to Ram Potham’s recent literature review and Ryan Greenblatt’s overview of areas in control work.

Duration: 6 weeks (1 session/week)
Approx dates: TBC
Time commitment: 2–5 hours/week
Format: Remote  (Weekly 2-hour discussion sessions + ~2 hours prep). You can find the curriculum structure here

Responsibilities:

• Lead weekly 2-hour discussions
• Explain technical concepts clearly and help connect ideas across the syllabus
• Encourage thoughtful participation and collaborative learning

Ideal background:

• Familiarity with at least 2–4 topics from Ram Potham’s AI control methods literature review
• Background in research, engineering, or technical AI safety (does not need to be in AI control specifically)

• Comfortable explaining technical concepts and facilitating discussion

   

Teaching Assistants

Program: Alignment Research Fellowship 

Duration: 8 weeks
Approx dates: TBC
Based on: ARENA curriculum

The Alignment Research Fellowship helps participants build technical foundations for alignment research, covering ML basics, transformers, and mechanistic interpretability. Final projects explore reinforcement learning, evaluation methods or mechanistic interpretability with NDIF.

Responsibilities:

• Guide participants through technical exercises
• Provide implementation feedback
• Help scope and supervise final projects

Ideal background:

• Strong PyTorch skills
• Familiarity with transformers and mechanistic interpretability
• Experience with RL or evaluations is a plus
• Prior participation in ARENA or similar programs is a plus

If you're unsure whether you're a good fit, err on the side of applying.

Why work with us?

Advance the field of AI safety: Contribute to a growing ecosystem of researchers working to reduce risks from transformative AI.
Give back: Help talented researchers access high-quality courses, enabling them to pursue impactful careers in AI safety.
Join a growing community: Connect with researchers focused on mitigating risks from transformative AI.

Other benefits include:

• Co-authorship on research papers (where applicable)
• Access to structured curricula and compute resources
• Letters of recommendation

Interested?
We welcome expressions of interest here or feel free to contact us at [email protected]