Claude Opus 4

Why AI safety teams are wrong to blame only alignment

AI models do not just fail from bad alignment; they also inherit harmful stories from training data.