What would make an RSP independently verifiable?
Read the current policy, pick one commitment, and ask what an outside researcher would need to evaluate it.
The gap between looking right and being right
ELK is a useful place to examine a deceptively simple question: what can an evaluator actually verify?
Start small. Leave something useful behind.
You do not need a complete theory of AI safety to make your next hour useful.
A shared evidence base, with uncertainty intact
The 2026 International AI Safety Report brings together more than 100 experts. Its assessment covers emerging risks and the evidence for managing them.
///
Good conversations start with a little curiosity.