← 研踪 PaperTrailACL Anthology · ACL 2026:acl:2026.acl-long.260 · 记录 #1
LLM-VA: Resolving the Jailbreak-Overrefusal Trade-off via Vector Alignment
来源日期（仅到月份） 2026-07
来源记录日期 2026-07
雷达观测 2026-11-01 22:04
摘要
Safety-aligned LLMs suffer from two failure modes: jailbreak (responding to harmful inputs) and over-refusal (declining benign queries). Existing vector steering methods adjust the magnitude of answer vectors, but this creates a fundamental trade-off—reducing jailbreak increases over-refusal and vice versa. We identify the root cause: LLMs encode the decision to respond (answer vector v_a) and the judgment of input safety (benign vector v_b) as nearly orthogonal directions, treating them as independent processes. We propose LLM-VA, which aligns v_a with v_b through closed-form weight updates, making the model’s willingness to respond causally dependent on its safety assessment—without fine-tuning or architectural changes. Our method identifies vectors at each layer using SVMs, selects safety-relevant layers, and iteratively aligns vectors via minimum-norm weight modifications. Experiments on 12 LLMs demonstrate that LLM-VA achieves 11.45% higher F1 than the best baseline while preserving 95.92% utility, and automatically adapts to each model’s safety bias without manual tuning.Code and models are available at https://hotbento.github.io/LLM-VA-Web/.
论文发现出处
这些页面提供论文线索，不代表同行评审或质量认证。
观测到的版本
平台记录号代表本站观测到的元数据变化，不是作者修订号。来源日期仅到月份或年份时，按该月或该年首日存档。
记录 #1 · 2026-07 · 最新观测版本
LLM-VA: Resolving the Jailbreak-Overrefusal Trade-off via Vector Alignment
Safety-aligned LLMs suffer from two failure modes: jailbreak (responding to harmful inputs) and over-refusal (declining benign queries). Existing vector steering methods adjust the magnitude of answer vectors, but this creates a fundamental trade-off—reducing jailbreak increases over-refusal and vice versa. We identify the root cause: LLMs encode the decision to respond (answer vector v_a) and the judgment of input safety (benign vector v_b) as nearly orthogonal directions, treating them as independent processes. We propose LLM-VA, which aligns v_a with v_b through closed-form weight updates, making the model’s willingness to respond causally dependent on its safety assessment—without fine-tuning or architectural changes. Our method identifies vectors at each layer using SVMs, selects safety-relevant layers, and iteratively aligns vectors via minimum-norm weight modifications. Experiments on 12 LLMs demonstrate that LLM-VA achieves 11.45% higher F1 than the best baseline while preserving 95.92% utility, and automatically adapts to each model’s safety bias without manual tuning.Code and models are available at https://hotbento.github.io/LLM-VA-Web/.
查看此版本解读 →查看此版本原文 ↗