Technical AI Safety Training
BASE’s AI Safety Training is a 16-week technical AI safety program based on the ARENA curriculum. The program combines remote classes with in-person instruction at FAR Labs in Berkeley and is designed to help participants build the technical skills and research experience needed to work on catastrophic risks from advanced AI systems.
The first 12 weeks focus on technical AI safety training, followed by a four-week capstone project. Participants receive mentorship, technical support, and access to computing resources while developing practical experience in areas such as interpretability, evaluations, and oversight.
After completing the 12 weeks of training, participants spend four weeks working on a capstone project focused on catastrophic AI risk. The capstone allows participants to apply what they have learned to a real-world research challenge while continuing to receive mentorship and technical support. The ARENA training program combines remote learning with in-person instruction at FAR Labs. It provides participants with access to teaching assistants, GPU and cloud computing resources, and other research tools needed for technical work.
From Training to Research
Who It's For
ARENA is intended for people who want to develop greater technical skills in AI safety and prepare for research or related roles in the field. Participants should already have some coding experience, particularly with Python, because the program includes hands-on technical exercises and research tasks. BASE is particularly focused on creating career pathways for Black people and others who remain underrepresented in technical AI safety.
Next Cohort in February 2027
Application open mid-Dec