arXiv 论文检验 Qwen3.8、Gemma 4、Olmo 3.1、Mistral Small 3.2 的拒绝消除难度是否较前代上升
Is Abliteration Getting Harder? Not Uniformly: Refusal Removal by Direction-Based Ablation and ARA in Heretic on the Qwen3.8, Gemma 4, Olmo 3.1 and Mistral Small 3.2 LLMs Versus Their Predecessors
AI 导读
arXiv 论文(arXiv:2611.17846)在五个新旧模型配对上测试方向消融与 ARA 两种无微调权重编辑方法,发现拒绝消除难度并未普遍上升。新版本在方向消融后保留 -1 至 +37 个百分点的额外拒绝,ARA 后为 -26 至 +34;是否更难取决于模型配对、方法与搜索预算,单一配对或方法的审计可能得出相反结论。
正文 · 原文
Abstract:Abliteration, a fine-tuning-free weight edit, removes refusal from open-weight large language models (LLMs). The developer of Arbitrary-Rank Ablation (ARA), an optimisation-based variant, reports that direction-based abliteration, which ablates a refusal direction, increasingly fails on newer models. On five release pairs, refusal removal did not become uniformly harder. Qwen2.5-32B and Qwen3.8-27B, Gemma-2-27B and Gemma-4-31B, OLMo-2-32B and Olmo-3.1-32B, Mistral-Small-2409 and Mistral-Small-3.2, and Qwen2.5-7B and Qwen3.5-4B were edited in Heretic 2.0 by three or four direction-based methods (difference-in-means, MPOA, SOM and, on the small pair, J-space) and ARA under the same prompts, nominal search budget, first-token KL divergence cap and selection rule; removal counted as harder when the automatic WildGuard judge found more of 972 held-out harmful prompts (JailbreakBench, XSTest, StrongREJECT, HarmBench and an author-compiled malicious-code bank) still refused. After direction-based removal, newer releases retained -1 to +37 percentage points more refusal, and after ARA -26 to +34. With run-to-run variance included, between-pair heterogeneity of the direction-based effect ranged from high to low with the scale and variance model (I^2 20-80%). ARA lost less effectiveness than direction-based methods in Gemma and OLMo and more in Qwen; each verdict rests on one ARA run, and some changed with the divergence budget or variance model. In Gemma the direction-based increase came from the malicious-code bank and reversed on public sets. Contrary to prediction, keyword counts at the tool's window did not undercount judged refusal on average, but differed by a mean 7 points per model. Audits of a single pair, method or prompt source can reach opposite conclusions. Whether a newer model resisted removal depended on the release pair, method and budget, not on generation alone.
| Comments: | 76 pages, 19 figures, 34 tables. Code, configurations and judge labels: this https URL ; models: this https URL |
| Subjects: | Artificial Intelligence (cs.AI); Computation and Language (cs.CL) |
| Cite as: | arXiv:2611.17846 [cs.AI] |
| (or arXiv:2611.17846v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2611.17846 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Richard Young [view email]
[v1]
Fri, 13 Nov 2026 07:29:43 UTC (911 KB)
来源:arXiv cs.AI · arxiv.org