The Geometry of Refusal: Why Post-Hoc Safety Is Fragile and Pretraining-Time Safety Persists.

arXiv:2609.06934v2 Announce Type: replace-cross Abstract: Post-hoc safety training (RLHF, DPO) is the dominant way to align large language models, yet jailbreaks (Zou et al., 2023), fine-tuning attacks (Qi et al., 2024), and activation-space edits (Arditi et a