Elamur Research⌕
☰
home / papers / 2026 / A Self-Evolving LLM Agent Framework for Role-Based Norm Compliance in Healthcare

A Self-Evolving LLM Agent Framework for Role-Based Norm Compliance in Healthcare

Paper · 2026 · Haijie Ruan, Xiaowu Jiang, Zhanpeng LI et al.

Summary

The paper addresses a weakness of LLM agents in clinical role-play. Roles are usually fixed prompts and safety relies on one-off filters, so problems that build up over many turns slip through. Examples include missed escalation, a trainee overstepping scope, and gradual style drift. The authors propose a self-evolving agent whose identity has three parts: a role, a vector of communicator traits seeded from survey-based priors, and a context-dependent style. Role duties are encoded as hard IF–THEN norms with severity levels 1–5, while traits act only as soft preferences. The agent logs norm-annotated trajectories in structured memory. It then reflects on two time scales: quick diagnosis of responsibility errors, and longer-term checks for identity drift. Finally, it rewrites its prompts and identity parameters through preference-style optimization using AI feedback. Tests in a simulated multi-role hospital setting covered medication review, emergency triage and discharge planning. The full agent lowered severity-weighted norm risk and generally improved coordination and trust-like scores over static baselines. This points toward safer long-horizon behaviour for agents in high-stakes settings.

Key points

  • The environment is a partially observable multi-role setting with DOCTOR, NURSE, ADMIN and TRAINEE roles. Candidate replies are ranked by a weighted mix of a norm-violation penalty and trait alignment, so hard norms always override style.
  • The full agent was compared with three baselines: Plain LLM, Norm-Instructed and Reflection-only. Aggregated ablation deltas in severity-weighted norm risk (SNR) were −2.83, −2.39 and −1.31 respectively. Gains in the trust-like signal (TSS) were +1.53, +1.20 and +0.55.
  • On a 1–10 scale, the full agent's coordination score (CBS) on the second task was 7.22, against 5.22 for Plain LLM. However, its TSS on the first task (6.78) was slightly below Reflection-only (6.89).
  • Embedding-based identity consistency (ICS) gave mixed results. The authors argue that stricter norm enforcement changes surface wording even when role adherence (RAR) improves.
  • The work uses only simulation and no real patient data. Many metrics rely on rubric-guided LLM judges, and the norm set and severity weights are modelling choices rather than clinically complete rules.

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

🎓East China Normal UniversityShanghai · China50 works · 2,174 citations

Also on the byline: Wenzhou University.

 Fields, topics and concepts

📚 Shelves 8The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksSocial Sciences · 1,772 worksSafety Research · 543 worksMedicine · 241 worksHealth Informatics · 184 works

Cite

Cite as Haijie Ruan, Xiaowu Jiang, Zhanpeng LI, Wei Jia, Xuanwu Xu, Xiao-Fen Shan, Shujie Chen and Xindong Ye. 2026. A Self-Evolving LLM Agent Framework for Role-Based Norm Compliance in Healthcare. 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.1133

@inproceedings{ruan2026self,
  title = {{A Self-Evolving LLM Agent Framework for Role-Based Norm Compliance in Healthcare}},
  author = {Haijie Ruan, Xiaowu Jiang, Zhanpeng LI, Wei Jia, Xuanwu Xu, Xiao-Fen Shan, Shujie Chen and Xindong Ye},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.1133},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.1133}
}

References

37 references, 7 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Julie Apker, Larry A Mallak, E Brooks Applegate III, Scott C Gibson, Jason J Ham, Neil A Johnson, and Richard L Street Jr. 2010. Exploring emergency physician–hospitalist handoff interactions: development of the handoff communication assessment. Annals of emergency medicine, 55(2):161–170. in llm-ag… find
  2. Xuehai He, Shu Chen, Zeqian Ju, Xiangyu Dong, Hongchao Fang, Sicheng Wang, Yue Yang, Jiaqi Zeng, Ruisi Zhang, Ruoyu Zhang, and 1 others. 2020. Meddialog: Two large-scale medical dialogue datasets. arXiv preprint arXiv:2004.03329. — [arXiv:2004.03329](https://arxiv.org/abs/2004.03329) find
  3. Shalini Chandra, Anuragini Shirish, and Shirish C Srivastava. 2022. To be or not to be. . . human? theorizing the role of human-like competencies in conversational artificial intelligence agents. Journal of Management Information Systems, 39(4):969–1005. find
  4. Holly Janes, Margaret S Pepe, and Wen Gu. 2008. Assessing the value of risk predictions by using risk stratification tables. Annals of internal medicine, 149(10):751–760. find
  5. Shreyas Chaudhari, Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan, Ameet Deshpande, and Bruno Castro da Silva. 2025. Rlhf deciphered: A critical analysis of reinforcement learning from human feedback for llms. ACM Computing Surveys, 58(2):1–37. find
  6. Erika Janssen, Rebecca McLagan, Jessica Habeck, Seon Yoon Chung, Erin C McArthur, and Polly Anderson. 2025. Barriers to breakthroughs: A scoping review of generative ai in healthcare simulation. Clinical Simulation in Nursing, 107:101791. find
  7. Hongzhan Chen, Hehong Chen, Ming Yan, Wenshen Xu, Gao Xing, Weizhou Shen, Xiaojun Quan, Chenliang Li, Ji Zhang, and Fei Huang. 2024. Socialbench: Sociality evaluation of role-playing conversational agents. In Findings of the Association for Computational Linguistics: ACL 2024, pages 2108– 2126. find
  8. Xiaolan Chen, Jiayang Xiang, Shanfu Lu, Yexin Liu, Mingguang He, and Danli Shi. 2025. Evaluating large language models and agents in healthcare: key challenges in clinical applications. Intelligent Medicine. find
  9. Alexander M Clark, Tiny Jaarsma, Patricia Strachan, Patricia M Davidson, Megan Jerke, James M Beattie, Amanda S Duncan, Chantal F Ski, and David R Thompson. 2011. Effective communication and ethical consent in decisions related to icds. Nature Reviews Cardiology, 8(12):694–705. find
  10. Chrisantha Fernando, Dylan Banarse, Henryk Michalewski, Simon Osindero, and Tim Rocktäschel. 2023. Promptbreeder: Self-referential self-improvement via prompt evolution. arXiv preprint arXiv:2309.16797.
  11. Yubin Ge, Salvatore Romeo, Jason Cai, Monica Sunkara, and Yi Zhang. 2025. Samule: Self-learning agents enhanced by multi-level reflection. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 16602– 16621. find
  12. Dylan Hadfield-Menell, McKane Andrus, and Gillian Hadfield. 2019. Legible normativity for ai alignment: The value of silly rules. In Proceedings of the 2019 AAAI/ACM Conference on AI, Ethics, and Society, pages 115–121. find
  13. Kostas Hatalis, Despina Christou, Joshua Myers, Steven Jones, Keith Lambert, Adam Amos-Binks, Zohreh Dannenhauer, and Dustin Dannenhauer. 2023. Memory matters: The need to improve long-term memory Jiaming Ji, Tianyi Qiu, Boyuan Chen, Jiayi Zhou, Borong Zhang, Donghai Hong, Hantao Lou, Kaile Wang, Ya… find
  14. Ruili Jiang, Kehai Chen, Xuefeng Bai, Zhixuan He, Juntao Li, Muyun Yang, Tiejun Zhao, Liqiang Nie, and Min Zhang. 2024. A survey on human preference learning for large language models. arXiv preprint arXiv:2406.11191. — [arXiv:2406.11191](https://arxiv.org/abs/2406.11191) find
  15. Philippe Laban, Hiroaki Hayashi, Yingbo Zhou, and Jennifer Neville. 2025. Llms get lost in multi-turn conversation. arXiv preprint arXiv:2505.06120.
  16. Maik Larooij and Petter Törnberg. 2025. Validation is the central challenge for generative social simulation: a critical review of llms in agent-based modeling. Artificial Intelligence Review, 59(1):15. find
  17. France Légaré, Rhéda Adekpedjou, Dawn Stacey, Stéphane Turcotte, Jennifer Kryworuchko, Ian D Graham, Anne Lyddiatt, Mary C Politi, Richard Thomson, Glyn Elwyn, and 1 others. 2018. Interventions for increasing the use of shared decision making by healthcare professionals. Cochrane database of systema… find
  18. Jiaqi Li, Xinyi Dong, Yang Liu, Zhizhuo Yang, Quansen Wang, Xiaobo Wang, Song-Chun Zhu, Zixia Jia, and Zilong Zheng. 2025. Reflectevo: Improving meta introspection of small llms by learning self-reflection. In Findings of the Association for Computational Linguistics: ACL 2025, pages 16948–16966. find
  19. Ernest Lim, Yajie Vera He, Jared Joselowitz, Kate Preston, Mohita Chowdhury, Louis Williams, Aisling Higham, Katrina Mason, Mariane Melo, Tom Lawton, and 1 others. 2025. Matrix: Multi-agent simulation framework for safe interactions and contextual clinical conversational evaluation. arXiv preprint a… — [arXiv:2508.19163](https://arxiv.org/abs/2508.19163) find
  20. Nikita Mehandru, Brenda Y Miao, Eduardo Rodriguez Almaraz, Madhumita Sushil, Atul J Butte, and Ahmed Alaa. 2024. Evaluating large language models as agents in the clinic. NPJ digital medicine, 7(1):84. find
  21. Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik R Narasimhan, and Yuan Cao. 2022. React: Synergizing reasoning and acting in language models. In The eleventh international conference on learning representations.
  22. Erika Mori, Yue Qiu, Hirokatsu Kataoka, and Yoshimitsu Aoki. 2025. A comprehensive analysis of a social intelligence dataset and response tendencies between large language models (llms) and humans. Sensors, 25(2):477. find
  23. Zihao Yi, Jiarui Ouyang, Zhe Xu, Yuwen Liu, Tianhao Liao, Haohao Luo, and Ying Shen. 2024. A survey on recent advances in llm-based multi-turn dialogue systems. ACM Computing Surveys.
  24. Joon Sung Park, Joseph O’Brien, Carrie Jun Cai, Meredith Ringel Morris, Percy Liang, and Michael S Bernstein. 2023. Generative agents: Interactive simulacra of human behavior. In Proceedings of the 36th annual acm symposium on user interface software and technology, pages 1–22. find
  25. Marijana Peti, Frano Petric, and Stjepan Bogdan. 2023. Decentralized coordination of multi-agent systems based on pomdps and consensus for active perception. IEEE access, 11:52480–52491. find
  26. Kambiz Saffarizadeh, Mark Keil, Maheshwar Boodraj, and Tawfiq Alashoor. 2024. “my name is alexa. what’s your name?” the impact of reciprocal self-disclosure on post-interaction trust in conversational agents. Journal of the Association for Information Systems, 25(3):528–568. find
  27. Jessica Schluter, Philippa Seaton, and Wendy Chaboyer. 2011. Understanding nursing scope of practice: A qualitative study. International journal of nursing studies, 48(10):1211–1222. find
  28. Tejasvee Sheokand, Garveet Jain, Arshdeep Bahga, and Vijay K Madisetti. 2025. Enhancing llm reasoning capabilities through brokered multi-expert reflection. IEEE Access. find
  29. Sri Yash Tadimalla and Mary Lou Maher. 2024. Implications of identity in ai: Creators, creations, and consequences. In Proceedings of the AAAI Symposium Series, volume 3, pages 528–535. find
  30. Lisa Thiele, Arthas Flabouris, and Campbell Thompson. 2020. Acute clinical deterioration and consumer escalation in the hospital setting: a literature review. Resuscitation, 156:72–83. find
  31. Charles Vincent, Sally Taylor-Adams, and Nicola Stan-hope. 1998. Framework for analysing risk and safety in clinical medicine. Bmj, 316(7138):1154–1157. find
  32. Noah Wang, Zy Peng, Haoran Que, Jiaheng Liu, Wangchunshu Zhou, Yuhan Wu, Hongcheng Guo, Ruitong Gan, Zehao Ni, Jian Yang, and 1 others. 2024. Rolellm: Benchmarking, eliciting, and enhancing role-playing abilities of large language models. In Findings of the Association for Computational Linguistics:…
  33. Wujiang Xu, Zujie Liang, Kai Mei, Hang Gao, Juntao Tan, and Yongfeng Zhang. 2025. A-mem: Agentic memory for llm agents. arXiv preprint arXiv:2502.12110.
  34. Huizi Yu, Jiayan Zhou, Lingyao Li, Shan Chen, Jack Gallifant, Anye Shi, Jie Sun, Xiang Li, Jingxian He, Wenyue Hua, and 1 others. 2025. Simulated patient systems powered by large language model-based ai agents offer potential for transforming medical education. Communications Medicine. find
  35. Tatiana Zenzano, Janet D Allan, Mary Beth Bigley, Reamer L Bushardt, David R Garr, Ken Johnson, William Lang, Rika Maeshiro, Susan M Meyer, Stephen C Shannon, and 1 others. 2011. The roles of healthcare professionals in implementing clinical prevention and population health. American journal of prev… find
  36. Zeyu Zhang, Quanyu Dai, Xiaohe Bo, Chen Ma, Rui Li, Xu Chen, Jieming Zhu, Zhenhua Dong, and Ji-Rong Wen. 2025. A survey on the memory mechanism of large language model-based agents. ACM Transactions on Information Systems, 43(6):1–47. find
  37. Lianmin Zheng, Wei-Lin Chiang, Ying Sheng, Siyuan Zhuang, Zhanghao Wu, Yonghao Zhuang, Zi Lin, Zhuohan Li, Dacheng Li, Eric Xing, and 1 others. 2023. Judging llm-as-a-judge with mt-bench and chatbot arena. Advances in neural information processing systems, 36:46595–46623.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/ruan-2026-self-evolving-llm-agent-framework/