Metraipolis › archive › dc_cbae92e324b6
Broken Symmetry in LLM Refusal: Answer Release Is More Local Than Refusal Restoration — arXiv:2608.15772
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2608.15772
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-18T17:30:11.142Z
- SHA-256 of the archived text
be82c08b0b16f45c223fc4cf41c4eb5b5643e72760c0d4c2a65a3605c2f54a78- Stable URL of this page
https://metraipolis.com/documento/dc_cbae92e324b6
How to cite
Broken Symmetry in LLM Refusal: Answer Release Is More Local Than Refusal Restoration — arXiv:2608.15772 — arXiv (en), https://arxiv.org/abs/2608.15772 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_cbae92e324b6 (Metraipolis archive dc_cbae92e324b6, sha256 be82c08b0b16f45c223fc4cf41c4eb5b5643e72760c0d4c2a65a3605c2f54a78)
BibTeX
@misc{dccbae92e324b6,
title = {Broken Symmetry in LLM Refusal: Answer Release Is More Local Than Refusal Restoration — arXiv:2608.15772},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2608.15772},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_cbae92e324b6; sha256 be82c08b0b16f45c223fc4cf41c4eb5b5643e72760c0d4c2a65a3605c2f54a78},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.15772. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
When a language model refuses to answer a prompt, it is unclear whether the correct answer is erased from its internal representations, or merely suppressed at the output layer. We investigate this mechanism using a controlled withhold setting, which yields perfectly matched answering and refusal trajectories for bidirectional activation patching. We uncover a causal asymmetry in intervention locality under matched causal interventions, which we term broken symmetry. Even when a model generates a clean refusal, the correct answer remains linearly recoverable from its hidden states. Furthermore, releasing this withheld answer is a highly local operation, requiring only a single-position patch. Conversely, the reverse operation is not equally local: reimposing suppression requires broader interventions across multiple positions, and assembling a coherent refusal sequence is more difficult still. We further demonstrate that while an average answer-to-refusal displacement vector marks the geometric difference between these states, it fails to act as a reliable, reversible linear control toggle between behaviours. Taken together, our findings show that refusal does not function as a simple symmetric switch. For safety and auditing, this implies that probe recoverability can overestimate true behavioural control, and locating refusal-relevant directions does not reliably grant the ability to steer a model from answering to coherent refusal. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-08-18 in cs.AI). Categorie arXiv: Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2608.15772 — 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 2026-10-18T16:29:13.829Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-08-18?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).