Metraipolis › archive › dc_8aeec034ec9d

When Safety Routing Breaks: Understanding Alignment Fragility under Benign Fine-Tuning — arXiv:2609.01455

Source: arXiv (en)
Original: https://arxiv.org/abs/2609.01455
License (as archived): CC0-1.0
Archived: 2026-10-17T06:34:22.320Z
SHA-256 of the archived text: 714d6b671d67c3d0c5500826fd609d8415d22ffc1a6a840e9c381a62f3f12160
Stable URL of this page: https://metraipolis.com/documento/dc_8aeec034ec9d

How to cite

When Safety Routing Breaks: Understanding Alignment Fragility under Benign Fine-Tuning — arXiv:2609.01455 — arXiv (en), https://arxiv.org/abs/2609.01455 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_8aeec034ec9d (Metraipolis archive dc_8aeec034ec9d, sha256 714d6b671d67c3d0c5500826fd609d8415d22ffc1a6a840e9c381a62f3f12160)

BibTeX

@misc{dc8aeec034ec9d,
  title = {When Safety Routing Breaks: Understanding Alignment Fragility under Benign Fine-Tuning — arXiv:2609.01455},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.01455},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_8aeec034ec9d; sha256 714d6b671d67c3d0c5500826fd609d8415d22ffc1a6a840e9c381a62f3f12160},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.01455. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Benign fine-tuning severely weakens the safety alignment of large language models (LLMs), so we study why refusal behavior is so fragile. While prior work often attributes this failure to gradient conflict, we propose a fundamentally different Fisher-geometric explanation: safety Fisher is low-rank, and alignment makes the safety geometry flatter while preserving an output-routing pathway. After 100 benign fine-tuning examples, this pathway is selectively re-sharpened in output-side MLP modules, explaining the asymmetric fragility: safety can collapse to high attack success rates, while general utility degrades mildly. The routing view also explains why few safety examples can restore refusal behavior, indicating that internal safety-relevant representations are preserved. Finally, we show that LoRA and ASAM mitigate early collapse by suppressing output-side sharpness, but their protection weakens at larger fine-tuning scales. Overall, safety failure is best understood as a disruption of a low-rank output-routing mechanism

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-02 in cs.AI).
Categorie arXiv: Cryptography and Security (cs.CR) ; Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2609.01455 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-17T06:15:12.900Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-02?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).