Sycophancy as a Multilingual Alignment Failure: How Safety Degrades Across Languages, Topics, and Models

• URL: http://arxiv.org/abs/2606.08451v1
• Date: Sun, 07 Jun 2026 04:50:40 GMT
• Title: Sycophancy as a Multilingual Alignment Failure: How Safety Degrades Across Languages, Topics, and Models
• Abstract summary: Safety-aligned large language models often exhibit sycophancy, which is the tendency to affirm users' opinions regardless of factual accuracy.<n>We present the first large-scale, multi-model evaluation of cross-lingual sycophancy, benchmarking textbfsix instruction-tuned models across textbf38 languages and textbf33 topic categories.
• Score: 3.2555924024638667
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Safety-aligned large language models often exhibit sycophancy, which is the tendency to affirm users' opinions regardless of factual accuracy. Although well-studied in English, its manifestation in other languages remains largely unexamined, leaving billions of non-English speakers potentially vulnerable to model-validated misinformation. We present the first large-scale, multi-model evaluation of cross-lingual sycophancy, benchmarking \textbf{six instruction-tuned models} across \textbf{1.1 million instances} spanning \textbf{38 languages} and \textbf{33 topic categories}. We identify a consistent resource-tier effect: sycophancy rates spike sharply in low-resource and zero-shot language settings. Critically, this degradation is topic-agnostic, as models fail uniformly across both benign and safety-critical prompts, offering no additional protection where it is most needed. We further identify tokenizer fertility as a structural driver of this alignment collapse. Collectively, our results demonstrate that prevailing alignment methodologies generalize poorly beyond high-resource languages, underscoring the urgent need for equitable multilingual safety techniques.

Related papers

• Language-Specific Gaps in AI Safety Training Datasets [0.0]
  Self-harm and sexual-content categories have no native-language coverage in either African-language tier we studied.<n>We show that self-harm and sexual-content categories have no native-language coverage in either African-language tier we studied.
  arXiv  Detail & Related papers  (2026-08-13T18:40:22Z)
• LLM Safety Alignment in Low-Resource Languages: A Systematic Literature Review [9.895109697164283]
  Large Language Models (LLMs) have achieved substantial progress in safety alignment.<n>Their safety guarantees remain significantly weaker in low-resource and multilingual settings.
  arXiv  Detail & Related papers  (2026-07-20T21:11:00Z)
• The Heterogeneous Safety Impacts of Benign Multilingual Fine-Tuning [13.578131927232292]
  Fine-tuning a large language model can increase its tendency to respond to unsafe adversarial prompts.<n>We present the first comprehensive empirical study of this phenomenon in multilingual settings by fine-tuning Llama-3.2, Qwen3, and Gemma-3 models.<n>We find that safety outcomes are highly sensitive to both the choice of fine-tuning language and the evaluation language.
  arXiv  Detail & Related papers  (2026-06-27T10:05:36Z)
• Why Do Safety Guardrails Degrade Across Languages? [21.521293656854183]
  Large language models exhibit safety degradation in non-English languages.<n>We introduce a latent variable model, a Multi-Group Item Response Theory (IRT) framework, that decouples safety-driving factors.<n>We evaluate the safety robustness of 61 model configurations across 5 closed-model families and 10 languages of varying resource.
  arXiv  Detail & Related papers  (2026-05-16T22:08:54Z)
• Multilingual Refusal Alignment for Safer Large Language Models [53.64286756804503]
  We investigate whether single-language alignment transfers cross-lingually, how language consistency is preserved during training, and the resulting trade-offs with general knowledge capabilities.<n>We introduce RefusEU, a novel refusal alignment dataset covering 12 European languages, including a dedicated test set for evaluating current state-of-the-art models.<n>Our controlled Direct Preference Optimization (DPO) experiments provide two key insights: aligning models exclusively in English is insufficient to ensure cross-lingual safety, even for the same harm categories, whereas training on multilingual datasets can improve safety without degrading general performance, as measured by the Global MMLU benchmark.
  arXiv  Detail & Related papers  (2026-04-24T09:29:14Z)
• All Languages Matter: Understanding and Mitigating Language Bias in Multilingual RAG [87.74635133954975]
  We show that current Multilingual Retrieval-Augmented Generation (mRAG) systems suffer from a language bias during reranking.<n>We propose textittextbfLanguage-textbfAgnostic textbfUtility-driven textbfReranker textbfAlignment (LAURA), which aligns multilingual evidence ranking with downstream generative utility.
  arXiv  Detail & Related papers  (2026-04-22T05:33:06Z)
• LASA: Language-Agnostic Semantic Alignment at the Semantic Bottleneck for LLM Safety [88.98698230989186]
  Large language models (LLMs) often demonstrate strong safety performance in high-resource languages, yet exhibit severe vulnerabilities in low-resource languages.<n>We attribute this gap to a mismatch between language-agnostic semantic understanding ability and language-dominant safety alignment biased toward high-resource languages.<n>We propose Language-Agnostic Semantic Alignment (LASA), which anchors safety alignment directly in semantic bottlenecks.
  arXiv  Detail & Related papers  (2026-04-13T15:59:50Z)
• Lost in Translation? A Comparative Study on the Cross-Lingual Transfer of Composite Harms [0.5376203747548287]
  Most safety evaluations of large language models (LLMs) remain anchored in English.<n>Some types of harm survive translation almost intact, while others distort or disappear.<n>We introduce CompositeHarm, a translation-based benchmark to examine how safety alignment holds up as both syntax and semantics shift.
  arXiv  Detail & Related papers  (2026-02-08T13:22:50Z)
• Lingua-SafetyBench: A Benchmark for Safety Evaluation of Multilingual Vision-Language Models [54.10540442330978]
  Existing benchmarks are typically multilingual but text-only, or multimodal but monolingual.<n>Recent multilingual red-teaming efforts render harmful prompts into images, yet rely heavily on typography-style visuals.<n>We introduce a benchmark of 100,440 harmful image-text pairs across 10 languages, explicitly partitioned into image-dominant and text-dominant subsets.
  arXiv  Detail & Related papers  (2026-01-30T09:18:13Z)
• Conditions for Catastrophic Forgetting in Multilingual Translation [24.10629800866219]
  We identify the conditions that trigger catastrophic forgetting in multilingual fine-tuning.<n>We show that the relative scale between model and data size is a primary determinant of forgetting.<n>We also show that cross-lingual alignment can mitigate forgetting while also facilitating positive transfer to unseen target languages.
  arXiv  Detail & Related papers  (2025-10-22T12:54:00Z)
• Beyond the Final Layer: Intermediate Representations for Better Multilingual Calibration in Large Language Models [50.34755385896279]
  Confidence calibration is crucial for the reliable deployment of Large Language Models (LLMs)<n>We conduct the first large-scale, systematic studies of multilingual calibration across six model families and over 100 languages.<n>We find that non-English languages suffer from systematically worse calibration.
  arXiv  Detail & Related papers  (2025-10-03T16:07:15Z)
• MrGuard: A Multilingual Reasoning Guardrail for Universal LLM Safety [56.77103365251923]
  Large Language Models (LLMs) are susceptible to adversarial attacks such as jailbreaking.<n>This vulnerability is exacerbated in multilingual settings, where multilingual safety-aligned data is often limited.<n>We introduce a multilingual guardrail with reasoning for prompt classification.
  arXiv  Detail & Related papers  (2025-04-21T17:15:06Z)
• The Hidden Space of Safety: Understanding Preference-Tuned LLMs in Multilingual context [0.9130277390156759]
  Alignment tuning has enabled large language models to excel in reasoning, instruction-following, and minimizing harmful generations.<n>Despite their widespread deployment, these models exhibit a monolingual bias, raising concerns about the effectiveness of alignment across languages.<n>Current alignment methods predominantly focus on English, leaving it unclear how alignment mechanism generalizes to multilingual settings.
  arXiv  Detail & Related papers  (2025-04-03T15:46:46Z)
• Adversarial GLUE: A Multi-Task Benchmark for Robustness Evaluation of Language Models [86.02610674750345]
  Adversarial GLUE (AdvGLUE) is a new multi-task benchmark to explore and evaluate the vulnerabilities of modern large-scale language models under various types of adversarial attacks. We apply 14 adversarial attack methods to GLUE tasks to construct AdvGLUE, which is further validated by humans for reliable annotations. All the language models and robust training methods we tested perform poorly on AdvGLUE, with scores lagging far behind the benign accuracy.
  arXiv  Detail & Related papers  (2021-11-04T12:59:55Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.