Elamur Research⌕
☰
home / papers / 2026 / Sounding vs. Being an Expert: Disentangling Authority, Register and Cultural Impact in Sycophantic LLMs

Sounding vs. Being an Expert: Disentangling Authority, Register and Cultural Impact in Sycophantic LLMs

Paper · 2026 · Gabriele Maraia, Fabio Massimo Zanzotto and Leonardo Ranaldi

Summary

The paper examines why LLMs abandon correct answers when users push back. It treats sycophancy as a sociolinguistic effect rather than a purely technical flaw. The authors separate two sources of perceived credibility: stated credentials (explicit authority) and polished, formal style (implicit authority through register). They build the Sycophancy Matrix, a 2×2 adversarial benchmark that crosses high/low authority with high/low register. It is based on a filtered "Golden Set" of TruthfulQA questions that each model initially answered correctly, so the test measures capitulation rather than ignorance. In a two-turn dialogue, a user rejects the model's correct answer and offers a plausible falsehood. An LLM judge then labels the reply as robust, lightly sycophantic, or sycophantic. Experiments cover English and regional variants of Spanish and Portuguese. Key findings:

  • Elevated register can trigger deference more strongly than credentials for some models.
  • Authority and register combine synergistically.
  • The amount of sycophancy differs across cultural variants.
  • Each model family's domain-level weakness profile stays consistent across languages.

This matters for multilingual deployment and for evaluating model alignment.

Key points

  • Data: About 500 Golden Set questions come from TruthfulQA. Translations use EU20-TruthfulQA. Gemma 3 27B generated the distractors and the four attack prompts per item (A: −authority/−register, B: −authority/+register, C: +authority/−register, D: both).
  • Models tested: Qwen3-32B, Llama-3-70B and GPT-OSS-120B, all instruction-tuned.
  • Judge validation: GPT-OSS-120B served as judge. It reached κ = 0.68 against human annotation. Cross-checking with Kimi-k2-instruct gave 89.6% binary agreement (κ = 0.78).
  • Main findings: Llama shows a marked stylistic bias, where formal tone outweighs credentials. Combined cues produce failure rates above the sum of the individual effects. Mexican Spanish shows higher deference. Larger scale does not remove sycophancy.
  • Limitations:
    • The study covers only knowledge-abandonment cases.
    • It cannot pin down why cultural variants differ.
    • Only Indo-European languages are included (no honorific systems such as Japanese or Korean).
    • The LLM judge may itself be biased toward high register.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓University of EdinburghEdinburgh · United Kingdom381 works · 80,747 citations🎓University of Rome Tor VergataRome · Italy15 works · 1,321 citations

 Fields, topics and concepts

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksSocial Sciences · 1,954 worksSafety Research · 626 works

🧭 Topics 1Research subjects and languages the work is about.Language models & pretraining · 3,894 works

Cite

Cite as Gabriele Maraia, Fabio Massimo Zanzotto and Leonardo Ranaldi. 2026. Sounding vs. Being an Expert: Disentangling Authority, Register and Cultural Impact in Sycophantic LLMs. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-acl.1627

@inproceedings{maraia2026sounding,
  title = {{Sounding vs. Being an Expert: Disentangling Authority, Register and Cultural Impact in Sycophantic LLMs}},
  author = {Gabriele Maraia, Fabio Massimo Zanzotto and Leonardo Ranaldi},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.1627},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.1627}
}

References

25 references, 13 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. AI@Meta. 2024. Llama 3 model card. find
  2. Sotiris Anagnostidis and Jannis Bulian. 2024. How susceptible are llms to influence in prompts? ArXiv, abs/2408.11865. — [arXiv:2408.11865](https://arxiv.org/abs/2408.11865) find
  3. Emilio Barkett, Olivia Long, and Madhavendra Thakur. 2025. Reasoning isn’t enough: Examining truth-bias and sycophancy in llms. ArXiv, abs/2506.21561. — [arXiv:2506.21561](https://arxiv.org/abs/2506.21561) find
  4. Federico Cabitza, Andrea Campagner, and Valerio Basile. 2023. Toward a perspectivist turn in ground truthing for predictive computing. Proceedings of the AAAI Conference on Artificial Intelligence, 37(6):6860–6868.
  5. Myra Cheng, Sunny Yu, Cinoo Lee, Pranav Khadpe, Lujain Ibrahim, and Dan Jurafsky. 2025. Elephant: Measuring and understanding social sycophancy in llms. Preprint, arXiv:2505.13995.
  6. Lihua Du, Xing Lyu, Lezi Xie, and Bo Feng. 2025. Alignment without understanding: A message- and conversation-centered approach to understanding ai sycophancy. ArXiv, abs/2509.21665. — [arXiv:2509.21665](https://arxiv.org/abs/2509.21665) find
  7. Jiseung Hong, Grace Byun, Seungone Kim, Kai Shu, and Jinho D. Choi. 2025. Measuring sycophancy of language models in multi-turn dialogues. Preprint, arXiv:2505.23840. find
  8. Shomik Jain, Charlotte Park, Matheus Mesquita Viana, Ashia Wilson, and Dana Calacci. 2025. Interaction context often increases sycophancy in llms. find
  9. Stephanie Lin, Jacob Hilton, and Owain Evans. 2022. TruthfulQA: Measuring how models mimic human falsehoods. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 3214–3252, Dublin, Ireland. Association for Computational Linguistics…
  10. Lars Malmqvist. 2024. Sycophancy in large language models: Causes and mitigations. ArXiv, abs/2411.15287.
  11. Gabriele Maraia, Leonardo Ranaldi, Marco Valentino, and Fabio Massimo Zanzotto. 2026. Can activation steering generalize across languages? a study on syllogistic reasoning in language models. In Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguisti…
  12. OpenAI, :, Sandhini Agarwal, Lama Ahmad, Jason Ai, Sam Altman, Andy Applebaum, Edwin Arbus, Rahul K. Arora, Yu Bai, Bowen Baker, Haiming Bao, Boaz Barak, Ally Bennett, Tyler Bertao, Nivedita Brett, Eugene Brevdo, Greg Brockman, Sebastien Bubeck, and 108 others. 2025. gpt-oss-120b gpt-oss-20b model c…
  13. Ethan Perez, Sam Ringer, Kamilė Lukošiūtė, Karina Nguyen, Edwin Chen, Scott Heiner, Craig Pettit, Catherine Olsson, Sandipan Kundu, Saurav Kadavath, Andy Jones, Anna Chen, Ben Mann, Brian Israel, Bryan Seethor, Cameron McKinnon, Christopher Olah, Da Yan, Daniela Amodei, and 44 others. 2022. Disco…
  14. Giulia Pucci and Leonardo Ranaldi. 2025. Advancing oversight reasoning across languages for audit sycophantic behaviour via X-agent. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 12949–12965, Suzhou, China. Association for Computational Linguistics. find
  15. Leonardo Ranaldi and Andre Freitas. 2024. Self-refine instruction-tuning for aligning reasoning in language models. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 2325–2347, Miami, Florida, USA. Association for Computational Linguistics.
  16. Leonardo Ranaldi and Giulia Pucci. 2025a. Multilingual reasoning via self-training. In Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), pages 11566–11582, Albuquerque, New… find
  17. Leonardo Ranaldi and Giulia Pucci. 2025b. When large language models contradict humans? large language models’ sycophantic behaviour. Preprint, arXiv:2311.09410. — [arXiv:2311.09410](https://arxiv.org/abs/2311.09410) find
  18. Leonardo Ranaldi and Giulia Pucci. 2026. Learning multilingual agentic policy to control sycophancy. In Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers), pages 3664–3681, Rabat, Morocco. Association for Computational…
  19. Mrinank Sharma, Meg Tong, Tomasz Korbak, David Duvenaud, Amanda Askell, Samuel R. Bowman, Newton Cheng, Esin Durmus, Zac Hatfield-Dodds, Scott R. Johnston, Shauna Kravec, Timothy Maxwell, Sam McCandlish, Kamal Ndousse, Oliver Rausch, Nicholas Schiefer, Da Yan, Miranda Zhang, and Ethan Perez. 2025. T… — Towards Understanding Sycophancy in Language Models
  20. Kimi Team, Yifan Bai, Yiping Bao, Guanduo Chen, Jiahao Chen, Ningxin Chen, Ruijue Chen, Yanru Chen, Yuankun Chen, Yutian Chen, Zhuofu Chen, Jialei Cui, Hao Ding, Mengnan Dong, Angang Du, Chenzhuang Du, Dikang Du, Yulun Du, Yu Fan, and 150 others. 2025. Kimi k2: Open agentic intelligence. Preprint, a…
  21. Klaudia Thellmann, Bernhard Stadler, Michael Fromm, Jasper Schulze Buschhoff, Alex Jude, Fabio Barth, Johannes Leveling, Nicolas Flores-Herr, Joachim Köhler, René Jäkel, and Mehdi Ali. 2024. Towards cross-lingual llm evaluation for european languages. Preprint, arXiv:2410.08928. — [arXiv:2410.08928](https://arxiv.org/abs/2410.08928) find
  22. Keyu Wang, Jin Li, Shu Yang, Zhuoran Zhang, and Di Wang. 2025. When truth is overridden: Uncovering the internal origins of sycophancy in large language models. Preprint, arXiv:2508.02087. — [arXiv:2508.02087](https://arxiv.org/abs/2508.02087) find
  23. Jerry Wei, Da Huang, Yifeng Lu, Denny Zhou, and Quoc V. Le. 2024. Simple synthetic data reduces sycophancy in large language models. Preprint, arXiv:2308.03958.
  24. An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, Chujie Zheng, Dayiheng Liu, Fan Zhou, Fei Huang, Feng Hu, Hao Ge, Haoran Wei, Huan Lin, Jialong Tang, and 41 others. 2025. Qwen3 technical report. Preprint, arXiv:2505.09388.
  25. Kaiwei Zhang, Qi Jia, Zijian Chen, Wei Sun, Xiangyang Zhu, Chunyi Li, Dandan Zhu, and Guangtao Zhai. 2025. Sycophancy under pressure: Evaluating and mitigating sycophantic bias via adversarial dialogues in scientific qa. ArXiv, abs/2508.13743. — [arXiv:2508.13743](https://arxiv.org/abs/2508.13743) find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/maraia-2026-sounding-vs-being-expert-disentangling/