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

arXiv: https://arxiv.org/abs/2606.08451 · HTML: https://mlsanity.com/paper/2606.08451/

## Key facts

- Cited by 1 papers in the map (data through 14 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.

## Placed in

- LLM alignment / RLHF / preference optimisation / [Safety alignment and jailbreak attacks](https://mlsanity.com/alignment/safety-alignment-and-jailbreak-attacks/)

## Builds on (most cited first, 23 of 23)

- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Unsupervised Cross-lingual Representation Learning at Scale](https://mlsanity.com/paper/1911.02116/) (2019)
- [mT5: A massively multilingual pre-trained text-to-text transformer](https://mlsanity.com/paper/2010.11934/) (2020)
- [RealToxicityPrompts: Evaluating Neural Toxic Degeneration in Language Models](https://mlsanity.com/paper/2009.11462/) (2020)
- [Towards Understanding Sycophancy in Language Models](https://mlsanity.com/paper/2310.13548/) (2023)
- [BeaverTails: Towards Improved Safety Alignment of LLM via a Human-Preference Dataset](https://mlsanity.com/paper/2307.04657/) (2023)
- [The State and Fate of Linguistic Diversity and Inclusion in the NLP World](https://mlsanity.com/paper/2004.09095/) (2020)
- [XTREME: A Massively Multilingual Multi-task Benchmark for Evaluating Cross-lingual Generalization](https://mlsanity.com/paper/2003.11080/) (2020)
- [Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training](https://mlsanity.com/paper/2401.05566/) (2024)
- [DecodingTrust: A Comprehensive Assessment of Trustworthiness in GPT Models](https://mlsanity.com/paper/2306.11698/) (2023)
- [How Good is Your Tokenizer? On the Monolingual Performance of Multilingual Language Models](https://mlsanity.com/paper/2012.15613/) (2020)
- [Low-Resource Languages Jailbreak GPT-4](https://mlsanity.com/paper/2310.02446/) (2023)
- [Do All Languages Cost the Same? Tokenization in the Era of Commercial Language Models](https://mlsanity.com/paper/2305.13707/) (2023)
- [Do-Not-Answer: A Dataset for Evaluating Safeguards in LLMs](https://mlsanity.com/paper/2308.13387/) (2023)
- [Safety Tax: Safety Alignment Makes Your Large Reasoning Models Less Reasonable](https://mlsanity.com/paper/2503.00555/) (2025)
- [Multilingual is not enough: BERT for Finnish](https://mlsanity.com/paper/1912.07076/) (2019)
- [Sycophancy in Large Language Models: Causes and Mitigations](https://mlsanity.com/paper/2411.15287/) (2024)
- [The Language Barrier: Dissecting Safety Challenges of LLMs in Multilingual Contexts](https://mlsanity.com/paper/2401.13136/) (2024)
- [Improving Multilingual Models with Language-Clustered Vocabularies](https://mlsanity.com/paper/2010.12777/) (2020)
- [Tower+: Bridging Generality and Translation Specialization in Multilingual LLMs](https://mlsanity.com/paper/2506.17080/) (2025)
- [Multilingual Machine Translation with Open Large Language Models at Practical Scale: An Empirical Study](https://mlsanity.com/paper/2502.02481/) (2025)
- [Difficulties with Evaluating a Deception Detector for AIs](https://mlsanity.com/paper/2511.22662/) (2025)
- [Align Once, Benefit Multilingually: Enforcing Multilingual Consistency for LLM Safety Alignment](https://mlsanity.com/paper/2602.16660/) (2026)

## Built on by (most cited first, 1 of 1)

- [Pushback in a Haystack: Language Models Respond to Factual Errors Ambiguously and Ineffectively](https://mlsanity.com/paper/2610.17703/) (2026)

## 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.