{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,11,11]],"date-time":"2026-11-11T21:40:04Z","timestamp":1794433204248,"version":"4.9.1"},"reference-count":48,"publisher":"Springer Science and Business Media LLC","issue":"10","license":[{"start":{"date-parts":[[2026,9,15]],"date-time":"2026-09-15T00:00:00Z","timestamp":1789430400000},"content-version":"tdm","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0"},{"start":{"date-parts":[[2026,9,15]],"date-time":"2026-09-15T00:00:00Z","timestamp":1789430400000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0"}],"content-domain":{"domain":["link.springer.com"],"crossmark-restriction":false},"short-container-title":["Nat Med"],"published-print":{"date-parts":[[2026,10]]},"abstract":"<jats:title>Abstract<\/jats:title>\n                  <jats:p>Autonomous clinical artificial intelligence (AI) agents powered by large language models (LLMs), meaning systems that can complete a diagnostic workflow without continuous human input, are increasingly capable of supporting complex reasoning and decision-making. Clinical translation, however, remains limited by two unmet requirements: institutionally governed deployment and reliable decision-time uncertainty estimation. Here we developed and evaluated a fully on-premise clinical agent that couples local operational control with a multi-perspective reliability framework to support selective autonomy. Across two Medical Information Mart for Intensive Care IV (MIMIC-IV)-derived benchmarks, the agent achieved 90.04% accuracy on a seven-disease task and 83.8% accuracy on a four-disease task, approaching a cloud baseline on the primary benchmark. To assess decision-time reliability, we quantified internal-likelihood, language-based and behavioral-stability measures for diagnosis and reasoning. Diagnostic behavioral consistency provided the strongest discrimination of correctness (area under the curve (AUC)\u2009=\u20090.860) and remained informative under stress testing (AUC\u2009=\u20090.875). At a consistency threshold of 0.90, 49.4% of cases were retained at 98.9% diagnostic accuracy. These findings support a practical framework for institutionally governed clinical agents in which decision-time reliability signals identify a lower-risk subset for autonomous handling and defer the remainder for review.<\/jats:p>","DOI":"10.1038\/s41591-026-04609-x","type":"journal-article","created":{"date-parts":[[2026,9,15]],"date-time":"2026-09-15T09:02:58Z","timestamp":1789462978000},"page":"3705-3716","update-policy":"https:\/\/doi.org\/10.1007\/springer_crossmark_policy","source":"Crossref","is-referenced-by-count":3,"title":["On-premise medical AI agents for reliable clinical decision-making"],"prefix":"10.1038","volume":"32","author":[{"ORCID":"https:\/\/orcid.org\/0009-0008-9965-8027","authenticated-orcid":false,"given":"Li","family":"Zhang","sequence":"first","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-0407-7617","authenticated-orcid":false,"given":"Georg","family":"W\u00f6lflein","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0009-0006-6195-9276","authenticated-orcid":false,"given":"Dyke","family":"Ferber","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Junhao","family":"Liang","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Zunamys I.","family":"Carrero","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-0447-0221","authenticated-orcid":false,"given":"Xuewei","family":"Wu","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Julien","family":"Vibert","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Jan","family":"Clusmann","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-4650-6827","authenticated-orcid":false,"given":"Lino","family":"M\u00f6hrmann","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Elena E.","family":"M\u00f6hrmann","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Catharina","family":"Wichmann","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Fabian","family":"Wolf","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Tim","family":"Lenz","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-3730-5348","authenticated-orcid":false,"given":"Jakob Nikolas","family":"Kather","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"297","published-online":{"date-parts":[[2026,9,15]]},"reference":[{"key":"4609_CR1","doi-asserted-by":"publisher","first-page":"499","DOI":"10.1038\/s41746-026-02674-7","volume":"9","author":"S Schmidgall","year":"2026","unstructured":"Schmidgall, S. et al. AgentClinic: a multimodal benchmark for tool-using clinical AI agents. npj Digit. Med. 9, 499 (2026).","journal-title":"npj Digit. Med."},{"key":"4609_CR2","doi-asserted-by":"publisher","first-page":"84","DOI":"10.1038\/s41746-024-01083-y","volume":"7","author":"N Mehandru","year":"2024","unstructured":"Mehandru, N. et al. Evaluating large language models as agents in the clinic. npj Digit. Med. 7, 84 (2024).","journal-title":"npj Digit. Med."},{"key":"4609_CR3","doi-asserted-by":"publisher","first-page":"159","DOI":"10.1038\/s41746-025-01550-0","volume":"8","author":"X Chen","year":"2025","unstructured":"Chen, X. et al. Enhancing diagnostic capability with multi-agents conversational large language models. npj Digit. Med. 8, 159 (2025).","journal-title":"npj Digit. Med."},{"key":"4609_CR4","doi-asserted-by":"publisher","first-page":"9","DOI":"10.1038\/s44387-025-00011-z","volume":"1","author":"S Zhou","year":"2025","unstructured":"Zhou, S. et al. Large language models for disease diagnosis: a scoping review. npj Artif. Intell. 1, 9 (2025).","journal-title":"npj Artif. Intell."},{"key":"4609_CR5","doi-asserted-by":"publisher","first-page":"50","DOI":"10.1518\/hfes.46.1.50.30392","volume":"46","author":"JD Lee","year":"2004","unstructured":"Lee, J. D. & See, K. A. Trust in automation: designing for appropriate reliance. Hum. Factors 46, 50\u201380 (2004).","journal-title":"Hum. Factors"},{"key":"4609_CR6","doi-asserted-by":"publisher","unstructured":"Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile https:\/\/doi.org\/10.6028\/NIST.AI.600-1 (National Institute of Standards and Technology, 2024).","DOI":"10.6028\/NIST.AI.600-1"},{"key":"4609_CR7","unstructured":"Regulation (EU) 2025\/327 of the European Parliament and of the Council of 11 February 2025 on the European Health Data Space and amending Directive 2011\/24\/EU and Regulation (EU) 2024\/2847 (Text with EEA relevance) http:\/\/data.europa.eu\/eli\/reg\/2025\/327\/oj (2025)."},{"key":"4609_CR8","unstructured":"Artificial intelligence guidance for IG professionals https:\/\/digital.nhs.uk\/data-and-information\/information-governance\/guidance\/artificial-intelligence\/guidance-for-ig-professionals (NHS Digital, 2026)."},{"key":"4609_CR9","doi-asserted-by":"crossref","unstructured":"Dennst\u00e4dt, F., Hastings, J., Putora, P. M., Schmerder, M. & Cihoric, N. Implementing large language models in healthcare while balancing control, collaboration, costs and security. npj Digit. Med. 8, 143 (2025).","DOI":"10.1038\/s41746-025-01476-7"},{"key":"4609_CR10","doi-asserted-by":"publisher","first-page":"105658","DOI":"10.1016\/j.imavis.2025.105658","volume":"162","author":"S Lembo","year":"2025","unstructured":"Lembo, S., Barra, P., Di Biasi, L., Bouwmans, T. & Tortora, G. AI4RDD: artificial intelligence and rare disease diagnosis: a proposal to improve the anamnesis process. Image Vis. Comput. 162, 105658 (2025).","journal-title":"Image Vis. Comput."},{"key":"4609_CR11","doi-asserted-by":"publisher","unstructured":"Geng, J. et al. A survey of confidence estimation and calibration in large language models. in Proc. 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) (eds Duh, K., Gomez, H. & Bethard, S.) 6577\u22126595 https:\/\/doi.org\/10.18653\/v1\/2024.naacl-long.366 (Association for Computational Linguistics, 2024).","DOI":"10.18653\/v1\/2024.naacl-long.366"},{"key":"4609_CR12","unstructured":"May a HIPAA covered entity or business associate use a cloud service to store or process ePHI? https:\/\/www.hhs.gov\/hipaa\/for-professionals\/faq\/2075\/may-a-hipaa-covered-entity-or-business-associate-use-cloud-service-to-store-or-process-ephi\/index.html (US Department of Health and Human Services, 2023)."},{"key":"4609_CR13","unstructured":"Regulation (EU) 2016\/679 of the European Parliament and of the Council of 27 April 2016 on the protection of natural persons with regard to the processing of personal data and on the free movement of such data, and repealing Directive 95\/46\/EC (General Data Protection Regulation) (Text with EEA relevance) http:\/\/data.europa.eu\/eli\/reg\/2016\/679\/oj (2016)."},{"key":"4609_CR14","unstructured":"Personal Information Protection and Electronic Documents Act. https:\/\/www.laws-lois.justice.gc.ca\/eng\/acts\/P-8.6\/ (Government of Canada, 2000)."},{"key":"4609_CR15","unstructured":"Marketing Submission Recommendations for a Predetermined Change Control Plan for Artificial Intelligence-Enabled Device Software Functions: Guidance for Industry and Food and Drug Administration Staff https:\/\/www.fda.gov\/regulatory-information\/search-fda-guidance-documents\/marketing-submission-recommendations-predetermined-change-control-plan-artificial-intelligence (US Food and Drug Administration, 2025)."},{"key":"4609_CR16","doi-asserted-by":"publisher","first-page":"924","DOI":"10.1038\/s41591-022-01772-9","volume":"28","author":"B Vasey","year":"2022","unstructured":"Vasey, B. et al. Reporting guideline for the early-stage clinical evaluation of decision support systems driven by artificial intelligence: DECIDE-AI. Nat. Med. 28, 924\u2013933 (2022).","journal-title":"Nat. Med."},{"key":"4609_CR17","unstructured":"Clinical Decision Support Software: Guidance for Industry and Food and Drug Administration Staff https:\/\/www.fda.gov\/regulatory-information\/search-fda-guidance-documents\/clinical-decision-support-software (US Food and Drug Administration, 2026)."},{"key":"4609_CR18","unstructured":"Applying Human Factors and Usability Engineering to Medical Devices: Guidance for Industry and Food and Drug Administration Staff https:\/\/www.fda.gov\/regulatory-information\/search-fda-guidance-documents\/applying-human-factors-and-usability-engineering-medical-devices (US Food and Drug Administration, 2016)."},{"key":"4609_CR19","unstructured":"Ethics and governance of artificial intelligence for health: WHO guidance https:\/\/www.who.int\/publications\/i\/item\/9789240029200 (World Health Organization, 2021)."},{"key":"4609_CR20","unstructured":"Anadkat, S. How to make your completions outputs consistent with the new seed parameter https:\/\/developers.openai.com\/cookbook\/examples\/reproducible_outputs_with_the_seed_parameter (2023)."},{"key":"4609_CR21","doi-asserted-by":"crossref","unstructured":"Zhao, Q. et al. Uncertainty Propagation on LLM Agent. in Proc. 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) 6064\u20136073 (Association for Computational Linguistics, 2025).","DOI":"10.18653\/v1\/2025.acl-long.302"},{"key":"4609_CR22","doi-asserted-by":"publisher","first-page":"1873","DOI":"10.1038\/s41591-025-03626-6","volume":"31","author":"M Omar","year":"2025","unstructured":"Omar, M. et al. Sociodemographic biases in medical decision making by large language models. Nat. Med. 31, 1873\u20131881 (2025).","journal-title":"Nat. Med."},{"key":"4609_CR23","unstructured":"Guo, C., Pleiss, G., Sun, Y. & Weinberger, K. Q. On calibration of modern neural networks. in International Conference on Machine Learning 1321\u22121330 https:\/\/proceedings.mlr.press\/v70\/guo17a.html (PMLR, 2017)."},{"key":"4609_CR24","unstructured":"Xiong, M. et al. Can LLMs express their uncertainty? An empirical evaluation of confidence elicitation in LLMs. in International Conference on Learning Representations (2024)."},{"key":"4609_CR25","doi-asserted-by":"publisher","first-page":"139","DOI":"10.1093\/jamia\/ocae254","volume":"32","author":"T Savage","year":"2025","unstructured":"Savage, T. et al. Large language model uncertainty proxies: discrimination and calibration for medical diagnosis and treatment. J. Am. Med. Inform. Assoc. 32, 139\u2013149 (2025).","journal-title":"J. Am. Med. Inform. Assoc."},{"key":"4609_CR26","doi-asserted-by":"publisher","first-page":"625","DOI":"10.1038\/s41586-024-07421-0","volume":"630","author":"S Farquhar","year":"2024","unstructured":"Farquhar, S., Kossen, J., Kuhn, L. & Gal, Y. Detecting hallucinations in large language models using semantic entropy. Nature 630, 625\u2013630 (2024).","journal-title":"Nature"},{"key":"4609_CR27","doi-asserted-by":"publisher","first-page":"690","DOI":"10.1038\/s41746-025-02071-6","volume":"8","author":"S Zhou","year":"2025","unstructured":"Zhou, S. et al. Uncertainty-aware large language models for explainable disease diagnosis. npj Digit. Med. 8, 690 (2025).","journal-title":"npj Digit. Med."},{"key":"4609_CR28","doi-asserted-by":"publisher","unstructured":"Taubenfeld, A. et al. Confidence improves self-consistency in LLMs. in Findings of the Association for Computational Linguistics: ACL 2025 (eds Che, W. et al.) 20090\u221220111 https:\/\/doi.org\/10.18653\/v1\/2025.findings-acl.1030 (Association for Computational Linguistics, 2025).","DOI":"10.18653\/v1\/2025.findings-acl.1030"},{"key":"4609_CR29","doi-asserted-by":"publisher","unstructured":"Ferber, D. et al. Towards autonomous medical artificial intelligence agents. Nature 655, 1282\u22121291 https:\/\/doi.org\/10.1038\/s41586-026-10675-5 (2026).","DOI":"10.1038\/s41586-026-10675-5"},{"key":"4609_CR30","doi-asserted-by":"publisher","first-page":"2613","DOI":"10.1038\/s41591-024-03097-1","volume":"30","author":"P Hager","year":"2024","unstructured":"Hager, P. et al. Evaluation and mitigation of the limitations of large language models in clinical decision-making. Nat. Med. 30, 2613\u20132622 (2024).","journal-title":"Nat. Med."},{"key":"4609_CR31","unstructured":"MIMIC CDM Leaderboard - a Hugging Face Space by MIMIC-CDM https:\/\/huggingface.co\/spaces\/MIMIC-CDM\/leaderboard"},{"key":"4609_CR32","doi-asserted-by":"crossref","unstructured":"Chiu, C., Pitis, S. & van der Schaar, M. Simulating viva voce examinations to evaluate clinical reasoning in large language models. in Advances in Neural Information Processing Systems 38 (2026).","DOI":"10.52202\/085713-4269"},{"key":"4609_CR33","doi-asserted-by":"publisher","unstructured":"Kissling, C., Merdjanovska, E. & Akbik, A. Self-aware knowledge probing: evaluating language models\u2019 relational knowledge through confidence calibration. Preprint at https:\/\/doi.org\/10.48550\/arXiv.2601.18901 (2026).","DOI":"10.48550\/arXiv.2601.18901"},{"key":"4609_CR34","unstructured":"Health Level Seven International. FHIR Specification https:\/\/hl7.org\/fhir\/"},{"key":"4609_CR35","doi-asserted-by":"publisher","unstructured":"GLM-4.5 Team. GLM-4.5: agentic, reasoning, and coding (ARC) foundation models. Preprint at https:\/\/doi.org\/10.48550\/arXiv.2508.06471 (2025).","DOI":"10.48550\/arXiv.2508.06471"},{"key":"4609_CR36","doi-asserted-by":"publisher","unstructured":"OpenAI. gpt-oss-120b & gpt-oss-20b model card. Preprint at https:\/\/doi.org\/10.48550\/arXiv.2508.10925 (2025).","DOI":"10.48550\/arXiv.2508.10925"},{"key":"4609_CR37","doi-asserted-by":"publisher","unstructured":"GLM-5-Team. GLM-5: from vibe coding to agentic engineering. Preprint at https:\/\/doi.org\/10.48550\/arXiv.2602.15763 (2026).","DOI":"10.48550\/arXiv.2602.15763"},{"key":"4609_CR38","unstructured":"Qwen Team. Qwen\/Qwen3.5-397B-A17B-FP8 \u00b7 Hugging Face https:\/\/huggingface.co\/Qwen\/Qwen3.5-397B-A17B-FP8"},{"key":"4609_CR39","doi-asserted-by":"publisher","unstructured":"Kwon, W. et al. Efficient memory management for large language model serving with PagedAttention. in Proc. 29th Symposium on Operating Systems Principles 611\u2013626 https:\/\/doi.org\/10.1145\/3600006.3613165 (ACM, Koblenz Germany, 2023).","DOI":"10.1145\/3600006.3613165"},{"key":"4609_CR40","unstructured":"OpenAI. OpenAI Agents SDK. https:\/\/github.com\/openai\/openai-agents-python"},{"key":"4609_CR41","unstructured":"Gemini 3.1 Flash-Lite Preview. Gemini API https:\/\/ai.google.dev\/gemini-api\/docs\/models\/gemini-3.1-flash-lite-preview"},{"key":"4609_CR42","unstructured":"Wang, X. et al. Self-consistency improves chain-of-thought reasoning in language models. in International Conference on Learning Representations (2023)."},{"key":"4609_CR43","doi-asserted-by":"publisher","first-page":"S9","DOI":"10.1186\/1471-2105-9-S11-S9","volume":"9","author":"V Vincze","year":"2008","unstructured":"Vincze, V., Szarvas, G., Farkas, R., M\u00f3ra, G. & Csirik, J. The BioScope corpus: biomedical texts annotated for uncertainty, negation and their scopes. BMC Bioinformatics 9, S9 (2008).","journal-title":"BMC Bioinformatics"},{"key":"4609_CR44","first-page":"321","volume":"2012","author":"DA Hanauer","year":"2012","unstructured":"Hanauer, D. A. et al. Hedging their mets: the use of uncertainty terms in clinical documents and its potential implications when sharing the documents with patients. AMIA Annu. Symp. Proc. 2012, 321\u2013330 (2012).","journal-title":"AMIA Annu. Symp. Proc."},{"key":"4609_CR45","doi-asserted-by":"publisher","first-page":"D267","DOI":"10.1093\/nar\/gkh061","volume":"32","author":"O Bodenreider","year":"2004","unstructured":"Bodenreider, O. The Unified Medical Language System (UMLS): integrating biomedical terminology. Nucleic Acids Res. 32, D267\u2013D270 (2004).","journal-title":"Nucleic Acids Res."},{"key":"4609_CR46","doi-asserted-by":"publisher","unstructured":"Neumann, M., King, D., Beltagy, I. & Ammar, W. ScispaCy: fast and robust models for biomedical natural language processing. in Proc. 18th BioNLP Workshop and Shared Task (eds Demner-Fushman, D. et al.) 319\u2212327 https:\/\/doi.org\/10.18653\/v1\/W19-5034 (Association for Computational Linguistics, 2019).","DOI":"10.18653\/v1\/W19-5034"},{"key":"4609_CR47","unstructured":"spaCy: industrial-strength natural language processing in Python https:\/\/spacy.io\/"},{"key":"4609_CR48","doi-asserted-by":"publisher","unstructured":"Zhang, L. CAPA. Zenodo https:\/\/doi.org\/10.5281\/zenodo.20233449 (2026).","DOI":"10.5281\/zenodo.20233449"}],"container-title":["Nature Medicine"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/www.nature.com\/articles\/s41591-026-04609-x.pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/www.nature.com\/articles\/s41591-026-04609-x","content-type":"text\/html","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/www.nature.com\/articles\/s41591-026-04609-x.pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,10,15]],"date-time":"2026-10-15T04:32:10Z","timestamp":1792038730000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.nature.com\/articles\/s41591-026-04609-x"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,9,15]]},"references-count":48,"journal-issue":{"issue":"10","published-print":{"date-parts":[[2026,10]]}},"alternative-id":["4609"],"URL":"https:\/\/doi.org\/10.1038\/s41591-026-04609-x","relation":{},"ISSN":["1078-8956","1546-170X"],"issn-type":[{"value":"1078-8956","type":"print"},{"value":"1546-170X","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,9,15]]},"assertion":[{"value":"10 February 2026","order":1,"name":"received","label":"Received","group":{"name":"ArticleHistory","label":"Article History"}},{"value":"27 July 2026","order":2,"name":"accepted","label":"Accepted","group":{"name":"ArticleHistory","label":"Article History"}},{"value":"15 September 2026","order":3,"name":"first_online","label":"First Online","group":{"name":"ArticleHistory","label":"Article History"}},{"value":"J.N.K. declares ongoing consulting services for AstraZeneca and Bioptimus. Furthermore, he holds shares in StratifAI, Synagen, Tremont AI, Saterra AI and Spira Labs and has received an institutional research grant from GlaxoSmithKline and AstraZeneca as well as honoraria from AstraZeneca, Bayer, Daiichi Sankyo, Eisai, Janssen, Merck, Merck Sharp & Dohme, Bristol Myers Squibb, Roche, Pfizer and Fresenius. D.F. holds shares of and is an employee at Synagen. T.L. declares consulting services for StratifAI. J.C. received honoraria from Johnson & Johnson. L.M. declares that his wife is an employee of Pfizer Pharma GmbH. E.E.M. is an employee of Pfizer Pharma GmbH. This work was conducted independently during parental leave, without funding or support from Pfizer. Pfizer had no role in the study. The other authors declare no competing interests.","order":1,"name":"Ethics","label":"Competing interests","group":{"name":"EthicsHeading","label":"Ethics"}}]}}