mlsanityData to 7 Oct 2026

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

arXiv 2606.08451 ↗First posted 7 Jun 2026cs.CL cs.AIcited by 0

On the map: LLM alignment / RLHF / preference optimisation → Safety alignment and jailbreak attacks

Abstract (arXiv)

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 six instruction-tuned models across 1.1 million instances spanning 38 languages and 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.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Safety alignment and jailbreak attacks, one of 10 directions in LLM alignment / RLHF / preference optimisation. Builds on 23 of its 37 references that are in the map. First posted on arXiv 7 Jun 2026.

Builds on

23 of 37 references.

Closest ancestors · 4

Same direction, newest first

  1. 2026Align Once, Benefit Multilingually: Enforcing Multilingual Consistency for LLM Safety Alignment18 Feb
  2. 2025Safety Tax: Safety Alignment Makes Your Large Reasoning Models Less Reasonable1 Mar
  3. 2024The Language Barrier: Dissecting Safety Challenges of LLMs in Multilingual Contexts23 Jan
  4. 2023BeaverTails: Towards Improved Safety Alignment of LLM via a Human-Preference Dataset10 Jul

Other references · 19

  1. 2022Training language models to follow instructions with human feedbackcited by 15,712
  2. 2019Unsupervised Cross-lingual Representation Learning at Scalecited by 3,907
  3. 2020mT5: A massively multilingual pre-trained text-to-text transformercited by 1,813
  4. 2020RealToxicityPrompts: Evaluating Neural Toxic Degeneration in Language Modelscited by 1,235
  5. 2023Towards Understanding Sycophancy in Language Modelscited by 988
  6. 2020The State and Fate of Linguistic Diversity and Inclusion in the NLP Worldcited by 789
  7. 2020XTREME: A Massively Multilingual Multi-task Benchmark for Evaluating Cross-lingual Generalizationcited by 738
  8. 2023DecodingTrust: A Comprehensive Assessment of Trustworthiness in GPT Modelscited by 505
  9. 2024Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Trainingcited by 478
  10. 2020How Good is Your Tokenizer? On the Monolingual Performance of Multilingual Language Modelscited by 313
  11. 2023Low-Resource Languages Jailbreak GPT-4cited by 290
  12. 2023Do All Languages Cost the Same? Tokenization in the Era of Commercial Language Modelscited by 186
7 more
  1. 2023Do-Not-Answer: A Dataset for Evaluating Safeguards in LLMscited by 176
  2. 2019Multilingual is not enough: BERT for Finnishcited by 118
  3. 2024Sycophancy in Large Language Models: Causes and Mitigationscited by 97
  4. 2020Improving Multilingual Models with Language-Clustered Vocabulariescited by 53
  5. 2025Tower+: Bridging Generality and Translation Specialization in Multilingual LLMscited by 40
  6. 2025Multilingual Machine Translation with Open Large Language Models at Practical Scale: An Empirical Studycited by 38
  7. 2025Difficulties with Evaluating a Deception Detector for AIscited by 11

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025The Hidden Space of Safety: Understanding Preference-Tuned LLMs in Multilingual context
  2. 2023Simple synthetic data reduces sycophancy in large language models
  3. 2026The Wrong Kind of Right: Quantifying and Localizing Misfired Alignment in LLMs
  4. 2025Measuring Sycophancy of Language Models in Multi-turn Dialogues
  5. 2025Safety Alignment Depth in Large Language Models: A Markov Chain Perspective
  6. 2026Multilingual Refusal Alignment for Safer Large Language Models
  7. 2025Alignment and Safety in Large Language Models: Safety Mechanisms, Training Paradigms, and Emerging Challenges
  8. 2026Are Aligned Large Language Models Still Misaligned?