Elamur Research⌕
☰
home / papers / 2026 / Chameleons and Guardians: Unveiling the Divergence in Personality Plasticity and Cognitive Resistance across LLMs

Chameleons and Guardians: Unveiling the Divergence in Personality Plasticity and Cognitive Resistance across LLMs

Paper · 2026 · Jiaxin Wu, Jie Zhao, Huangxinrong et al.

Summary

The paper asks whether the personality an LLM shows on a psychometric test can be deliberately reshaped, and whether models differ in how easily this happens. The authors ran MBTI tests on seven recent commercial models and found that nearly all default to ENFJ. A plain persona instruction shifts Western models (Gemini, Claude, GPT) but barely moves Eastern ones (Qwen, DeepSeek, Kimi).

To overcome this resistance, they build the Personality Induction Framework (PIF). A Generator agent rewrites MBTI items into varied, context-rich statements. A Judge agent checks each one and requests revisions until it no longer resembles the original question. The shuffled outputs then serve as inducing contexts, and their strength can be tuned separately for each dimension.

PIF moves models toward target types. However, the resistant "Guardian" models respond to maximum (100%) intensity by swinging away from the target, an effect the authors call "Alignment Rebound." The findings matter for assigning roles in multi-agent systems and for understanding the tension between obeying instructions and keeping a stable identity.

Key points

  • Baseline test: Models tested were Gemini3-Pro, Claude-Opus-4.5, GPT-5.2, Qwen3-Max, DeepSeek-V3.2-Chat, Doubao-Seed-1.6flash and Kimi2. Each was run at temperatures 0.001, 0.5 and 1, with 20 trials per setting. Most stayed ENFJ regardless of temperature, while Gemini3 showed several types.
  • Two groups of models:
    • "Chameleons" (mostly Western models) show high plasticity under persona prompts.
    • "Guardians" (mostly Eastern models) resist change.
    • Some types, such as ISFP, were nearly impossible to induce in either group using standard prompts.
  • How PIF works: It relies on Generator–Judge collaboration, iterative refinement and shuffled paraphrases. These steps aim to ensure that personality shifts reflect genuine adaptation rather than recall of memorized test items. Intensity control allows fine-grained modulation along each MBTI dimension.
  • Downstream task (Social Utility Annotation): Personality-shifted models produced markedly different annotations.
    • DeepSeek accuracy rose from 0.7280 up to 0.9360, and its F1 from 0.4688 up to 0.7273.
    • Claude's F1 rose from 0.5106 up to 0.6667.
  • Generalization: PIF also changed Big Five trait scores. It worked with Chinese prompts as well: Claude transformed fully, while DeepSeek resisted on some dimensions.

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

🎓Guangdong University of TechnologyGuangzhou · China5 works · 112 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 Jiaxin Wu, Jie Zhao, Huangxinrong, Cuihong Zhang and Xiuzhu Wu. 2026. Chameleons and Guardians: Unveiling the Divergence in Personality Plasticity and Cognitive Resistance across 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.205

@inproceedings{wu2026chameleons,
  title = {{Chameleons and Guardians: Unveiling the Divergence in Personality Plasticity and Cognitive Resistance across LLMs}},
  author = {Jiaxin Wu, Jie Zhao, Huangxinrong, Cuihong Zhang and Xiuzhu Wu},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.205},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.205}
}

References

47 references, 9 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Lisa P. Argyle, Ethan C. Busby, Nancy Fulda, and Joshua R. Gubler. 2023. Out of one, many: Using language models to simulate human samples. Political Analysis, pages 337–351.
  2. S. Bama, M. S. Hema, S. Esakkirajan, and M. Nageswara Guptha. 2025. A hierarchical transformer network with label attention for personality prediction by mbti classification. Applied Soft Computing, page 113267. find
  3. Natalie A. Carlson and Vanessa Burbano. 2026. The use of llms to annotate data in management research: Foundational guidelines and warnings. Strategic Management Journal, pages 699–725. find
  4. Yun-Shiuan Chuang, Krirk Nirunwiroj, Studdiford Zach, Agam Goyal, Vincent V. Frigo, Sijia Yang, Dhavan V. Shah, Junjie Hu, and Timothy T. Rogers. 2024. Beyond demographics: Aligning role-playing llm-based agents using human belief networks. In Findings of the Association for Computational Linguistic… find
  5. João Coelho, Bruno Martins, João Magalhães, Jamie Callan, and Chenyan Xiong. 2024. Dwell in the beginning: How language models embed long documents for dense retrieval. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics, pages 370–377. find
  6. Marharyta Domnich, Julius Valja, Rasmus Moorits Veski, Giacomo Magnifico, Kadi Tulver, Eduard Barbu, and Raul Vicente. 2025. Towards unifying evaluation of counterfactual explanations: Leveraging large language models for human-centric assessments. In Proceedings of the AAAI Conference on Artificial… find
  7. Fahed Elourajini and Esma Aïmeur. 2022. Aws-ep: a multi-task prediction approach for mbti/big5 personality tests. In IEEE International Conference on Data Mining Workshops, pages 1–8. find
  8. Xiaoxu Han, Wanqing Zhao, Ziyu Guan, and Jinye Peng. 2025. Act-llm: A whole-process chain for character-centric role-playing with large language models. Expert Systems with Applications, page 129024. find
  9. Kai He, Yucheng Huang, Wenqing Wang, Delong Ran, Dongming Sheng, Junxuan Huang, Qika Lin, Jiaxing Xu, Wenqiang Liu, and Mengling Feng. 2025. Crab: A novel configurable role-playing llm with assessing benchmark. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistic… find
  10. Felix Hill, Roi Reichart, and Anna Korhonen. 2015. Simlex-999: Evaluating semantic models with (genuine) similarity estimation. Computational Linguistics, pages 665–695.
  11. Pranav Bhandari, Nicolas Fay, Michael J Wise, Amitava Datta, Meek Stephanie, Usman Naseem, and Mehwish Nasim. 2025. Can llm agents maintain a persona in discourse? In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, page 29201–29217. find
  12. Sirui Hong, Mingchen Zhuge, Jonathan Chen, Xiawu Zheng, Yuheng Cheng, Jinlin Wang, Ceyao Zhang, Zili Wang, Steven Ka Shing Yau, Zijuan Lin, Liyang Zhou, Chenyu Ran, Lingfeng Xiao, Chenglin Wu, and Jurgen Schmidhuber. 2023. Metagpt: Meta programming for a multi-agent collaborative framework. In Inter…
  13. Weihong Bi, Feifei Kou, Lei Shi, Yawen Li, Haisheng Li, Jinpeng Chen, and Mingying Xu. 2025. Leveraging the dual capabilities of llm: Llm-enhanced text mapping model for personality detection. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 23487–23495. find
  14. Sviatlana Höhn, Jauwairia Nasir, Daniel C. Tozadore, Ali Paikan Paikan, Pouyan Ziafati, and Elisabeth André. 2024. Beyond pretend-reality dualism: frame analysis of llm-powered role play with social agents. In Proceedings of the 12th International Conference on Human-Agent Interaction, pages 393–395… find
  15. Ke Ji, Yixin Lian, Linxu Li, Jingsheng Gao, Weiyuan Li, and Bin Dai. 2025. Enhancing persona consistency for llms’ role-playing using persona-aware contrastive learning. arXiv preprint arXiv:2503.17662. find
  16. Hang Jiang, Xiajie Zhang, Xubo Cao, Cynthia Breazeal, Deb Roy, and Jad Kabbara. 2024. Personallm: Investigating the ability of large language models to express personality traits. In Findings of the Association for Computational Linguistics: NAACL, page 3605–3627.
  17. Oliver P. John and Sanjay Srivastava. 1999. The Big Five Trait taxonomy: History, measurement, and theoretical perspectives. Handbook of Personality: Theory and Research. find
  18. Chuyi Kong, Ziyang Luo, Hongzhan Lin, Zhiyuan Fan, Yaxin Fan, Yuxi Sun, and Jing Ma. 2025. Sharp: Unlocking interactive hallucination via stance transfer in role-playing llms. In Findings of the Association for Computational Linguistics: ACL, pages 839–866. find
  19. Lucio La Cava and Andrea Tagarelli. 2025. Open models, closed minds? on agents capabilities in mimicking human personalities through open large language models. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 1355–1363. find
  20. Joint modeling of big five and questionnaire item-level scores. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 1456–1464. find
  21. Marilù Miotto, Nicola Rossberg, and Bennett Kleinberg. 2022. Who is gpt-3? an exploration of personality, values and demographics. In Proceedings of the Fifth Workshop on Natural Language Processing and Computational Social Science, pages 218–227. find
  22. Isabel Briggs Myers. 1962. The myers-briggs type indicator. Consulting Psychologists Press, Palo Alto, CA. find
  23. Isabel Briggs Myers. 1985. A guide to the development and use of the Myers-Briggs type indicator: Manual. Consulting Psychologists Press. find
  24. Mirabel Reid and Santosh S. Vempala. 2025. Does gpt really get it? a hierarchical scale to quantify human and ai’s understanding of algorithms. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 1492–1500. find
  25. Murray Shanahan, Kyle McDonell, and Laria Reynolds. 2023. Role play with large language models. Nature, pages 493–498.
  26. Yaniv Leviathan, Matan Kalman, and Yossi Matias. 2025. Prompt repetition improves non-reasoning llms. arXiv preprint arXiv:2512.14982. find
  27. Yunfan Shao, Linyang Li, Junqi Dai, and Xipeng Qiu. 2023. Character-llm: A trainable agent for role-playing. arXiv preprint arXiv:2310.10158.
  28. Bohan Li, Jiannan Guan, Longxu Dou, Yunlong Feng, Dingzirui Wang, Yang Xu, Enbo Wang, Qiguang Chen, Bichen Wang, xiao Xu, Yimeng Zhang, Libo Qin, Yanyan Zhao, Qingfu Zhu, and Wanxiang Che. 2025. Can large language models understand you better? an mbti personality detection dataset aligned with popul… find
  29. Haojun Shi, Suyu Ye, Xinyu Fang, Chuanyang Jin, Leyla Isik, Yen-Ling Kuo, and Tianmin Shu. 2025. Muma-tom: Multi-modal multi-agent theory of mind. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 1510–1519. find
  30. Wenhan Liu, Xinyu Ma, Yutao Zhu, Lixin Su, Shuaiqiang Wang, Dawei Yin, and Zhicheng Dou. 2025. Coranking: Collaborative ranking with small and large ranking agents. In Proceedings of the Conference on Empirical Methods in Natural Language Processing. find
  31. Tian Ma, Kaiyu Feng, Yu Rong, and Kangfei Zhao. 2025. From post to personality: Harnessing llms for mbti prediction in social media. In Proceedings of the 34th ACM International Conference on Information and Knowledge Management, pages 5011–5015. find
  32. Fanhang Man, Huandong Wang, Jianjie Fang, Zhaoyi Deng, Baining Zhao, Xinlei Chen, and Yong Li. 2025. Context-aware sentiment forecasting via llm-based multi-perspective role-playing agents. arXiv preprint arXiv:2505.24331. find
  33. Ryo Masumura, Shota Orihashi, Mana Ihori, Tomohiro Tanaka, Naoki Makishima, Satoshi Suzuki, Saki Mizuno, and Nobukatsu Hojo. 2025. Multimodal fine-grained apparent personality trait recognition: Kazutoshi Shinoda, Nobukatsu Hojo, Kyosuke Nishida, Saki Mizuno, Keita Suzuki, Ryo Masumura, Hiroaki Sugi… find
  34. Cheryl Siy, Yuanxin Pang, Kihoon Jung, and John Kim. 2025. One mbti does not fit all: Perceptions and usage of mbti in social media profiles. In Proceedings of the ACM on Human-Computer Interaction, pages 1–29. find
  35. Yuxuan Song, Qiudan Li, Yilin Wu, David Jingjun Xu, and Daniel Dajun Zeng. 2025. Knowledge-enhanced hierarchical heterogeneous graph for personality identification with limited training data. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 1529–1537. find
  36. Aleksandra Sorokovikova, Sharwin Rezagholi, Natalia Fedorova, and Ivan P. Yamshchikov. 2024. Llms simulate big5 personality traits: Further evidence. In Proceedings of the 1st Workshop on Personalization of Generative AI Systems, pages 83–87. find
  37. Jacob Mitchell Springer, Suhas Kotha, Daniel Fried, Graham Neubig, and Aditi Raghunathan. 2025. Repetition improves language model embeddings. In Proceedings of the International Conference on Learning Representations.
  38. Bin Tang, Ke-Qi Pan, Miao Zheng, Ning Zhou, Jia-Lu Sui, Dandan Zhu, Cheng-Long Deng, and Shu-Guang Kuai. 2025. Pose as a modality: A psychology-inspired network for personality recognition with a new multimodal dataset. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 1538– 15… find
  39. Yu-Min Tseng, Yu-Chao Huang, Teng-Yun Hsiao, Wei-Lin Chen, Chao-Wei Huang, Yu Meng, and Yun-Nung Chen. 2024. Two tales of persona in llms: A survey of role-playing and personalization. arXiv preprint arXiv:2406.01171.
  40. Lei Wang, Jianxun Lian, Yi Huang, Yanqi Dai, Haoxuan Li, Xu Chen, Xing Xie, and Ji-Rong Wen. 2025. Characterbox: Evaluating the role-playing capabilities of llms in text-based virtual worlds. In Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computat… find
  41. Noah Wang, Z.y. Peng, Haoran Que, Jiaheng Liu, Wangchunshu Zhou, Yuhan Wu, Hongcheng Guo, Ruitong Gan, Zehao Ni, Jian Yang, Man Zhang, Zhaoxiang Zhang, Wanli Ouyang, Ke Xu, Wenhao Huang, Jie Fu, and Junran Pneng. 2024. Rolellm: Benchmarking, eliciting, and enhancing role-playing abilities of large l…
  42. Qingyun Wu, Gagan Bansal, Jieyu Zhang, Yiran Wu, Beibin Li, Erkang Zhu, Li Jiang, Xiaoyun Zhang, Shaokun Zhang, Jiale Liu, Ahmed Hassan Awadallah, Ryen W White, Doug Burger, and Chi Wang. 2024a. Autogen: Enabling next-gen llm applications via multi-agent conversations. In First Conference on Languag… find
  43. Weiqi Wu, Hongqiu Wu, Lai Jiang, Xingyuan Liu, Hai Zhao, and Min Zhang. 2024b. From role-play to drama-interaction: An llm solution. In Findings of the Association for Computational Linguistics: ACL, page 3271–3290. find
  44. Migyeong Yang, Jiwon Kim, Minji Kim, and Jinyoung Han. 2025. What is your mbti?”: Predicting the personality types using hierarchical attention and graph learning. Expert Systems with Applications, page 129295. find
  45. Tao Yang, Tianyuan Shi, Fanqi Wan, Xiaojun Quan, Qifan Wang, Bingzhe Wu, and Jiaxiang Wu. 2023. Psycot: Psychological questionnaire as powerful chain-of-thought for personality detection. In Findings of the Association for Computational Linguistics: EMNLP, pages 3305–3320. find
  46. Dingbo Yuan, Yipeng Chen, Guodong Liu, Chenchen Li, Chengfu Tang, Dongxu Zhang, Zhenkui Wang, Xudong Wang, and Song Liu. 2025. Dmt-rolebench: A dynamic multi-turn dialogue based benchmark for role-playing evaluation of large language model and agent. In Proceedings of the AAAI Conference on Artifici… find
  47. Lili Zhao, Yang Wang, Qi Liu, Mengyun Wang, Wei Chen, Zhichao Sheng, and Shijin Wang. 2025. Evaluating large language models through role-guide and self-reflection: A comparative study. In International Conference on Learning Representations. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/wu-2026-chameleons-guardians-unveiling-divergence-personality/