AI safety and alignment in 2026: a crash course
A sourced crash course on AI safety: alignment, interpretability, evaluations, oversight, safety institutes and the 2026 frontier.
AI safety and alignment
Research on making AI systems reliable, interpretable and aligned with human intent, and how it is evaluated.
AI safety tracker: alignment, interpretability and evals in 2026
Live tracker of AI safety milestones: interpretability results, evaluations, safety frameworks, incidents and institutes.
AI control
CAISI / CAISSI: the US government's AI testing centre at NIST
Constitutional AI
Frontier safety frameworks (responsible scaling policies)
Reinforcement learning from human feedback (RLHF)
Scalable oversight
Sparse autoencoders (interpretability)
UK AI Security Institute (AISI)
Can AI safety keep pace with AI capabilities?
How AI safety testing works: evals, red teams and thresholds
How mechanistic interpretability works: looking inside AI
What is AI alignment? Making AI do what we intend