A Gravitational Interpretation of Safety Reversion under Fine-Tuning
Safety alignment in large language models can degrade during post-training even when neither the data nor the objective is intentionally adversarial. Alignment rebound and reverse dynamics suggest that this degradation may reactivate behavior suppressed during safety alignment. Building on these ideas, we hypothesize that ordinary non-adversarial post-training follows a reversion direction: the activation-space displacement from the safety-aligned model toward a more permissive, earlier helpful-only state. We see that for Llama, every tested trajectory across references, tasks, and seeds exceeds a matched empirical null, while at aligned Llama and Qwen checkpoints, a vocabulary readout sh...
| Authors | Samuele Poppi, Nils Lukas |
|---|---|
| Primary category | cs.LG |
| Published | 26 June 2026 |
| Last updated at source | 28 September 2026 |
Silytics shows public metadata from arXiv and links to the original. Information may be incomplete or out of date. Last refreshed 17 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.