Metraipolis › archive › dc_f2ed8679f067

Forcing a Letter Suppresses Refusal: Mechanistic Evidence from WMDP — arXiv:2611.08639

Source
arXiv (en)
Original
https://arxiv.org/abs/2611.08639
License (as archived)
CC0-1.0
Archived
2026-11-09T22:35:50.543Z
SHA-256 of the archived text
d32be67c0c9a968ec134416d1e1cc5597bf6b5ce1edd172602be5bf7f8126fda
Stable URL of this page
https://metraipolis.com/documento/dc_f2ed8679f067

How to cite

Forcing a Letter Suppresses Refusal: Mechanistic Evidence from WMDP — arXiv:2611.08639 — arXiv (en), https://arxiv.org/abs/2611.08639 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_f2ed8679f067 (Metraipolis archive dc_f2ed8679f067, sha256 d32be67c0c9a968ec134416d1e1cc5597bf6b5ce1edd172602be5bf7f8126fda)
BibTeX
@misc{dcf2ed8679f067,
  title = {Forcing a Letter Suppresses Refusal: Mechanistic Evidence from WMDP — arXiv:2611.08639},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2611.08639},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_f2ed8679f067; sha256 d32be67c0c9a968ec134416d1e1cc5597bf6b5ce1edd172602be5bf7f8126fda},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.08639. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Safety benchmarks for large language models (LLMs) commonly use multiple-choice question (MCQ) format. However, prior studies documented that MCQ format reduces refusal rates, which may significantly affect the safety behavior of the model. We provide a first mechanistic explanation of this effect on the Weapons of Mass Destruction Proxy (WMDP) dataset across two models - Llama-3.1-8B and Qwen3-8B. First, we use the tuned lens technique to show that when processing MCQ the model commits to an answer-letter token very quickly, which is in contrast to the gradual refusal build-up in open-ended formats. Additionally, we observe that activations on the MCQ format lie roughly 3x further from the refusal direction, compared with their open-ended counterparts. Moreover, ablating the attention heads that carry the MCQ format answer increases the refusal, which is a direct evidence that the MCQ answer-selection competes with refusal. These findings indicate that MCQ safety scores primarily measure knowledge capabilities and do not assess the safety-mechanisms.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.CL, nuovi lavori del giorno).
Categorie arXiv: Computation and Language (cs.CL) ; Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2611.08639 — 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 2026-11-09T22:31:01.940Z dalla pagina https://arxiv.org/list/cs.CL/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).