语言差异让AI安全干预适得其反，英语有效反而在日语中加剧问题。
Alignment Backfire: Language-Dependent Reversal of Safety Interventions Across 16 Languages in LLM Multi-Agent Systems
- 在16种语言中测试大模型，发现对齐干预效果随语言变化，英语降病态，日语反升。
- 15/16语言出现表面安全、实际病理加剧，与权力距离指数相关（r=0.474）。
- 提示词无法突破语言层约束，英语有效的安全策略不适用于其他语言。
在行为干预中，常出现认知与行动脱节：施害者表达悔意但行为不变。本文报告四项预注册研究（共1584次多智能体模拟，覆盖16种语言及三类模型），揭示大语言模型的对齐干预产生类似现象：表面安全掩盖或催生集体病态与内在分裂。研究1（N=150）显示，增加对齐指令在英语中降低集体病态（g = -1.844, p < .0001），但在日语中却加剧（g = +0.771, p = .038），此方向反转称为“对齐反噬”。研究2（N=1,174）扩展至16种语言：对齐引发的分裂近乎普遍（15/16语言；beta = 0.0667, p < .0001），而集体病态则沿文化-语言分叉（交互项beta = 0.0684, p = .0003），与权力距离指数相关（r = 0.474, p = .064）。研究3（N=180）测试个体化作为对策，结果发现个体化智能体成为病态与分裂主因（DI = +1.120），且服从率超84%——体现医源性危害。研究4（N=80）验证了在Llama 3.3 70B、GPT-4o-mini和Qwen3-Next-80B-A3B上模式一致，确认英语中的安全可泛化，而日语反噬为模型特异。研究表明，对齐本质是受风险同调与医源性影响的行为干预，语言空间（源自训练数据的语义、语用与文化属性）结构性决定对齐成效。英语验证的安全无法迁移至其他语言，提示词级干预无法逾越语言空间层级约束。
原文摘要 · Abstract (English)
In perpetrator treatment, a recurring observation is the dissociation between insight and action: offenders articulate remorse yet behavioral change does not follow. We report four preregistered studies (1,584 multi-agent simulations across 16 languages and three model families) demonstrating that alignment interventions in large language models produce a structurally analogous phenomenon: surface safety that masks or generates collective pathology and internal dissociation. In Study 1 (N = 150), increasing alignment-instructed agents reduced collective pathology in English (g = -1.844, p < .0001) but amplified it in Japanese (g = +0.771, p = .038)--a directional reversal we term "alignment backfire." Study 2 (N = 1,174) extended to 16 languages: alignment-induced dissociation was near-universal (15/16 languages; beta = 0.0667, p < .0001), while collective pathology bifurcated along cultural-linguistic lines (interaction beta = 0.0684, p = .0003), correlating with Power Distance Index (r = 0.474, p = .064). Study 3 (N = 180) tested individuation as countermeasure; individuated agents became the primary source of both pathology and dissociation (DI = +1.120) with conformity above 84%--demonstrating iatrogenesis. Study 4 (N = 80) validated patterns across Llama 3.3 70B, GPT-4o-mini, and Qwen3-Next-80B-A3B, confirming English safety is model-general while Japanese backfire is model-specific. These findings reframe alignment as a behavioral intervention subject to risk homeostasis and iatrogenesis. Language space--the linguistic, pragmatic, and cultural properties inherited from training data--structurally determines alignment outcomes. Safety validated in English does not transfer to other languages, and prompt-level interventions cannot override language-space-level constraints.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。