{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"institution":[{"name":"bioRxiv"}],"indexed":{"date-parts":[[2026,9,21]],"date-time":"2026-09-21T18:09:14Z","timestamp":1790014154499,"version":"4.0.1"},"posted":{"date-parts":[[2026,9,5]]},"group-title":"Bioinformatics","reference-count":59,"publisher":"openRxiv","license":[{"start":{"date-parts":[[2026,9,5]],"date-time":"2026-09-05T00:00:00Z","timestamp":1788566400000},"content-version":"vor","delay-in-days":0,"URL":"http:\/\/creativecommons.org\/licenses\/by-nc-nd\/4.0\/"}],"funder":[{"name":"NIA","award":["AG066515"],"award-info":[{"award-number":["AG066515"]}]},{"name":"NIA","award":["AG089509"],"award-info":[{"award-number":["AG089509"]}]},{"name":"NIA","award":["AG066206"],"award-info":[{"award-number":["AG066206"]}]},{"name":"NIA","award":["AG066515"],"award-info":[{"award-number":["AG066515"]}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"accepted":{"date-parts":[[2026,9,16]]},"abstract":"<jats:title>Abstract<\/jats:title>\n                <jats:p>Sensitive human data underpin discoveries across medicine, biology and the social sciences, yet privacy regulation often prevents sharing them with collaborators or artificial intelligence (AI) systems. We introduce SPHERE, a model-free method that makes sensitive datasets directly usable by AI and shareable for open science as a synthetic twin, while the original records never leave the local environment. Across 33 datasets spanning five scientific domains, SPHERE protects individual privacy against adversarial re-identification attacks while preserving the data\u2019s statistical structure: means, variances and correlations are reproduced exactly, effect size and P value in linear statistical analysis are numerically identical, nonlinear machine-learning utility is retained, and each twin is generated in seconds on a laptop. Frontier AI agents running on the twin reach the same scientific conclusions as on the original records. Analyses of the twin reproduce genome- and proteome-wide results at UK Biobank scale and recover the findings of landmark studies across three independent cohorts and consortia. The approach also extends to deep-learning embeddings across language, vision and time-series, with minimal utility loss. We make the Stanford Alzheimer\u2019s Disease Research Center cohort openly available for the first time, as a SPHERE twin spanning nine modalities that any registered researcher can analyze without an approval process. We release SPHERE with certification of each twin\u2019s privacy and fidelity, and an AI agent that autonomously executes research tasks on sensitive data without ever accessing it. Sensitive datasets that are currently closed to research could thus become routine inputs to open science and AI to enable key discoveries.<\/jats:p>","DOI":"10.64898\/2026.09.01.748580","type":"posted-content","created":{"date-parts":[[2026,9,6]],"date-time":"2026-09-06T05:40:12Z","timestamp":1788673212000},"source":"Crossref","is-referenced-by-count":0,"title":["Unlocking Sensitive Data with SPHERE in the Age of AI"],"prefix":"10.64898","author":[{"given":"Zihuai","family":"He","sequence":"first","affiliation":[{"name":"Department of Neurology and Neurological Sciences, Stanford University School of Medicine, Stanford, CA 94305, USA"},{"name":"Quantitative Sciences Unit, Division of Computational Medicine, Stanford University School of Medicine, Stanford, CA 94305, USA"},{"name":"Department of Biomedical Data Science, Stanford University School of Medicine, Stanford, CA 94305, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Junyoung","family":"Park","sequence":"additional","affiliation":[{"name":"Department of Neurology and Neurological Sciences, Stanford University School of Medicine, Stanford, CA 94305, USA"},{"name":"Quantitative Sciences Unit, Division of Computational Medicine, Stanford University School of Medicine, Stanford, CA 94305, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Rafael Catoia","family":"Pulgrossi","sequence":"additional","affiliation":[{"name":"Quantitative Sciences Unit, Division of Computational Medicine, Stanford University School of Medicine, Stanford, CA 94305, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Justin","family":"Lee","sequence":"additional","affiliation":[{"name":"Quantitative Sciences Unit, Division of Computational Medicine, Stanford University School of Medicine, Stanford, CA 94305, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"suffix":"III","given":"Robert R","family":"Butler","sequence":"additional","affiliation":[{"name":"Department of Neurology and Neurological Sciences, Stanford University School of Medicine, Stanford, CA 94305, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Audrey","family":"Weber","sequence":"additional","affiliation":[{"name":"Department of Neurology and Neurological Sciences, Stanford University School of Medicine, Stanford, CA 94305, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Lu","family":"Tian","sequence":"additional","affiliation":[{"name":"Department of Biomedical Data Science, Stanford University School of Medicine, Stanford, CA 94305, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Xiang","family":"Zhang","sequence":"additional","affiliation":[{"name":"Department of Computer Science, University of North Carolina at Charlotte, NC 28223, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Julie Fangran","family":"Wang","sequence":"additional","affiliation":[{"name":"Institute for Computational and Mathematical Engineering, Stanford University, Stanford, CA 94305, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Sharon","family":"Sha","sequence":"additional","affiliation":[{"name":"Department of Neurology and Neurological Sciences, Stanford University School of Medicine, Stanford, CA 94305, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Elizabeth C.","family":"Mormino","sequence":"additional","affiliation":[{"name":"Department of Neurology and Neurological Sciences, Stanford University School of Medicine, Stanford, CA 94305, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Tony","family":"Wyss-Coray","sequence":"additional","affiliation":[{"name":"Department of Neurology and Neurological Sciences, Stanford University School of Medicine, Stanford, CA 94305, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Victor W.","family":"Henderson","sequence":"additional","affiliation":[{"name":"Department of Neurology and Neurological Sciences, Stanford University School of Medicine, Stanford, CA 94305, USA"},{"name":"Department of Epidemiology and Population Health, Stanford University School of Medicine, Stanford, CA 94305, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Frank M.","family":"Longo","sequence":"additional","affiliation":[{"name":"Department of Neurology and Neurological Sciences, Stanford University School of Medicine, Stanford, CA 94305, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"James","family":"Zou","sequence":"additional","affiliation":[{"name":"Department of Biomedical Data Science, Stanford University School of Medicine, Stanford, CA 94305, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Manisha","family":"Desai","sequence":"additional","affiliation":[{"name":"Quantitative Sciences Unit, Division of Computational Medicine, Stanford University School of Medicine, Stanford, CA 94305, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Russ","family":"Altman","sequence":"additional","affiliation":[{"name":"Department of Bioengineering, Stanford University, Stanford, CA 94305, USA"},{"name":"Department of Genetics, Stanford University, Stanford, CA 94305, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"54368","reference":[{"key":"2026092110251413000_2026.09.01.748580v2.1","doi-asserted-by":"publisher","DOI":"10.1016\/j.cell.2024.09.022"},{"key":"2026092110251413000_2026.09.01.748580v2.2","doi-asserted-by":"publisher","DOI":"10.1038\/s41586-023-05881-4"},{"key":"2026092110251413000_2026.09.01.748580v2.3","doi-asserted-by":"publisher","DOI":"10.1038\/s41586-023-06221-2"},{"key":"2026092110251413000_2026.09.01.748580v2.4","doi-asserted-by":"publisher","DOI":"10.1126\/science.adi1778"},{"key":"2026092110251413000_2026.09.01.748580v2.5","unstructured":"Hoffmann, J. et al. Training compute-optimal large language models. arXiv preprint arXiv:2203.15556 (2022)."},{"key":"2026092110251413000_2026.09.01.748580v2.6","doi-asserted-by":"publisher","DOI":"10.1038\/s41586-023-06792-0"},{"key":"2026092110251413000_2026.09.01.748580v2.7","doi-asserted-by":"publisher","DOI":"10.1038\/S41586-025-09442-9"},{"key":"2026092110251413000_2026.09.01.748580v2.8","first-page":"1","article-title":"Health Insurance Portability and Accountability Act of 1996","volume":"104","year":"1996","journal-title":"Public law"},{"key":"2026092110251413000_2026.09.01.748580v2.9","unstructured":"General Data Protection Regulation (GDPR). Intersoft Consulting, Accessed in October 24 (2018)."},{"key":"2026092110251413000_2026.09.01.748580v2.10","doi-asserted-by":"crossref","first-page":"1539","DOI":"10.1038\/s41431-025-01927-5","article-title":"MINDDS-connect: a federated data platform integrating biobanks for meta cohort building and analysis","volume":"33","year":"2025","journal-title":"Eur. J. Hum. Genet"},{"key":"2026092110251413000_2026.09.01.748580v2.11","first-page":"1","article-title":"Advances and open problems in federated learning","volume":"14","year":"2021","journal-title":"Found. Trends Mach. Learn"},{"key":"2026092110251413000_2026.09.01.748580v2.12","doi-asserted-by":"publisher","DOI":"10.1038\/s41588-020-0651-0"},{"key":"2026092110251413000_2026.09.01.748580v2.13","doi-asserted-by":"publisher","DOI":"10.1038\/s41467-019-10933-3"},{"key":"2026092110251413000_2026.09.01.748580v2.14","doi-asserted-by":"crossref","first-page":"2892","DOI":"10.1016\/j.csbj.2024.07.005","article-title":"Synthetic data generation methods in healthcare: A review on open-source tools and methods","volume":"23","year":"2024","journal-title":"Comput. Struct. Biotechnol. J"},{"key":"2026092110251413000_2026.09.01.748580v2.15","doi-asserted-by":"crossref","unstructured":"Dwork, C. , McSherry, F. , Nissim, K. & Smith, A. Calibrating noise to sensitivity in private data analysis. in Theory of cryptography conference. 265\u2013284 (Springer, 2006).","DOI":"10.1007\/11681878_14"},{"key":"2026092110251413000_2026.09.01.748580v2.16","unstructured":"Xu, L. , Skoularidou, M. , Cuesta-Infante, A. & Veeramachaneni, K. Modeling tabular data using conditional gan. Adv. Neural Inf. Process. Syst. 32 (2019)."},{"key":"2026092110251413000_2026.09.01.748580v2.17","unstructured":"Kotelnikov, A. , Baranchuk, D. , Rubachev, I. & Babenko, A. TabDDPM: modelling tabular data with diffusion models. in International conference on machine learning. 17564\u201317579 (PMLR, 2023)."},{"key":"2026092110251413000_2026.09.01.748580v2.18","doi-asserted-by":"publisher","DOI":"10.18637\/JSS.V074.I07"},{"key":"2026092110251413000_2026.09.01.748580v2.19","doi-asserted-by":"publisher","DOI":"10.56553\/popets-2023-0055"},{"key":"2026092110251413000_2026.09.01.748580v2.20","doi-asserted-by":"publisher","DOI":"10.1561\/0400000042"},{"key":"2026092110251413000_2026.09.01.748580v2.21","first-page":"86","article-title":"Random orthogonal matrix masking methodology for microdata release","volume":"2","year":"2008","journal-title":"Int. J. Inf. Comput. Secur"},{"key":"2026092110251413000_2026.09.01.748580v2.22","unstructured":"Zhang, H. , Cisse, M. , Dauphin, Y. N. & Lopez-Paz, D. mixup: Beyond empirical risk minimization. arXiv preprint arXiv:1710.09412 (2017)."},{"key":"2026092110251413000_2026.09.01.748580v2.23","doi-asserted-by":"crossref","unstructured":"Patki, N. , Wedge, R. & Veeramachaneni, K. The synthetic data vault. in 2016 IEEE international conference on data science and advanced analytics (DSAA). 399\u2013410 (IEEE, 2016).","DOI":"10.1109\/DSAA.2016.49"},{"key":"2026092110251413000_2026.09.01.748580v2.24","doi-asserted-by":"crossref","unstructured":"Akinbami, L. et al. National Health and Nutrition Examination Survey, 2017\u2013March 2020 prepandemic file: sample design, estimation, and analytic guidelines. (2022).","DOI":"10.15620\/cdc:115434"},{"key":"2026092110251413000_2026.09.01.748580v2.25","doi-asserted-by":"publisher","DOI":"10.1161\/CIR.0000000000000596"},{"key":"2026092110251413000_2026.09.01.748580v2.26","doi-asserted-by":"publisher","DOI":"10.1126\/science.1095019"},{"key":"2026092110251413000_2026.09.01.748580v2.27","doi-asserted-by":"publisher","DOI":"10.1038\/nmeth.3746"},{"key":"2026092110251413000_2026.09.01.748580v2.28","doi-asserted-by":"publisher","DOI":"10.1038\/s41598-023-34866-6"},{"key":"2026092110251413000_2026.09.01.748580v2.29","doi-asserted-by":"crossref","unstructured":"Greshake, K. et al. Not what you\u2019ve signed up for: compromising real-world LLM-integrated applications with indirect prompt injection. in Proceedings of the 16th ACM Workshop on Artificial Intelligence and Security 79\u201390 (ACM, 2023).","DOI":"10.1145\/3605764.3623985"},{"key":"2026092110251413000_2026.09.01.748580v2.30","doi-asserted-by":"crossref","first-page":"e2549963","DOI":"10.1001\/jamanetworkopen.2025.49963","article-title":"Vulnerability of large language models to prompt injection when providing medical advice","volume":"8","year":"2025","journal-title":"JAMA Netw. Open"},{"key":"2026092110251413000_2026.09.01.748580v2.31","first-page":"80079","article-title":"Jailbroken: How does llm safety training fail?","volume":"36","year":"2023","journal-title":"Adv. Neural Inf. Process. Syst"},{"key":"2026092110251413000_2026.09.01.748580v2.32","unstructured":"Zou, A. , et al. Universal and transferable adversarial attacks on aligned language models. arXiv preprint arXiv:2307.15043 (2023)."},{"key":"2026092110251413000_2026.09.01.748580v2.33","doi-asserted-by":"crossref","unstructured":"Zharmagambetov, A. et al. AgentDAM: Privacy leakage evaluation for autonomous web agents. Adv. Neural Inf. Process. Syst. 38 (2025).","DOI":"10.52202\/085713-4609"},{"key":"2026092110251413000_2026.09.01.748580v2.34","doi-asserted-by":"crossref","first-page":"2556","DOI":"10.1038\/s41591-025-03834-0","article-title":"The Global Neurodegeneration Proteomics Consortium: biomarker and drug target discovery for common neurodegenerative diseases and aging","volume":"31","year":"2025","journal-title":"Nat. Med"},{"key":"2026092110251413000_2026.09.01.748580v2.35","doi-asserted-by":"crossref","first-page":"2703","DOI":"10.1038\/s41591-025-03798-1","article-title":"Plasma proteomics links brain and immune system aging with healthspan and longevity","volume":"31","year":"2025","journal-title":"Nat. Med"},{"key":"2026092110251413000_2026.09.01.748580v2.36","doi-asserted-by":"publisher","DOI":"10.64898\/2026.08.14.26360394"},{"key":"2026092110251413000_2026.09.01.748580v2.37","doi-asserted-by":"crossref","unstructured":"Morris, J. , Kuleshov, V. , Shmatikov, V. & Rush, A. M. Text embeddings reveal (almost) as much as text. in Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing. 12448\u201312460 (Association for Computational Linguistics, 2023).","DOI":"10.18653\/v1\/2023.emnlp-main.765"},{"key":"2026092110251413000_2026.09.01.748580v2.38","doi-asserted-by":"crossref","unstructured":"Huang, Y.-H. , Tsai, Y. , Hsiao, H. , Lin, H.-Y. & Lin, S.-D. Transferable embedding inversion attack: uncovering privacy risks in text embeddings without model queries. in Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 4193\u20134205 (Association for Computational Linguistics, 2024).","DOI":"10.18653\/v1\/2024.acl-long.230"},{"key":"2026092110251413000_2026.09.01.748580v2.39","unstructured":"Ouaari, S. , Kreuer, J. & Pfeifer, N. How private are DNA embeddings? Inverting foundation model representations of genomic sequences. arXiv preprint arXiv:2603.06950 (2026)."},{"key":"2026092110251413000_2026.09.01.748580v2.40","doi-asserted-by":"publisher","DOI":"10.1109\/TPAMI.2018.2827389"},{"key":"2026092110251413000_2026.09.01.748580v2.41","doi-asserted-by":"crossref","unstructured":"Ni, J. et al. Large dual encoders are generalizable retrievers. in Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing 9844\u20139855 (Association for Computational Linguistics, 2022).","DOI":"10.18653\/v1\/2022.emnlp-main.669"},{"key":"2026092110251413000_2026.09.01.748580v2.42","doi-asserted-by":"crossref","unstructured":"He, K. , Zhang, X. , Ren, S. & Sun, J. Deep residual learning for image recognition. in 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 770\u2013778 (IEEE, 2016).","DOI":"10.1109\/CVPR.2016.90"},{"key":"2026092110251413000_2026.09.01.748580v2.43","doi-asserted-by":"crossref","unstructured":"Turcan, E. & McKeown, K. Dreaddit: a Reddit dataset for stress analysis in social media. in Proceedings of the Tenth International Workshop on Health Text Mining and Information Analysis (LOUHI 2019) 97\u2013107 (Association for Computational Linguistics, 2019).","DOI":"10.18653\/v1\/D19-6213"},{"key":"2026092110251413000_2026.09.01.748580v2.44","doi-asserted-by":"crossref","unstructured":"Deng, J. , et al. ImageNet: a Large-Scale Hierarchical Image Database. (2009).","DOI":"10.1109\/CVPR.2009.5206848"},{"key":"2026092110251413000_2026.09.01.748580v2.45","doi-asserted-by":"publisher","DOI":"10.1787\/df1031a4-en"},{"key":"2026092110251413000_2026.09.01.748580v2.46","doi-asserted-by":"publisher","DOI":"10.1111\/bjet.13224"},{"key":"2026092110251413000_2026.09.01.748580v2.47","doi-asserted-by":"publisher","DOI":"10.4230\/LIPIcs.AFT.2023.12"},{"key":"2026092110251413000_2026.09.01.748580v2.48","unstructured":"Zhu, L. , Liu, Z. & Han, S. Deep leakage from gradients. Adv. Neural Inf. Process. Syst. 32 (2019)."},{"key":"2026092110251413000_2026.09.01.748580v2.49","unstructured":"Bischl, B. et al. OpenML benchmarking suites. in Proceedings of the Neural Information Processing Systems Track on Datasets and Benchmarks (NeurIPS Datasets and Benchmarks 2021) (2021)."},{"key":"2026092110251413000_2026.09.01.748580v2.50","first-page":"507","article-title":"Why do tree-based models still outperform deep learning on typical tabular data?","volume":"35","year":"2022","journal-title":"Adv. Neural Inf. Process. Syst"},{"key":"2026092110251413000_2026.09.01.748580v2.51","unstructured":"Fischer, S. F. , Harutyunyan, L. , Feurer, M. & Bischl, B. OpenML-CTR23 \u2013 a curated tabular regression benchmarking suite. in AutoML Conference 2023 (Workshop Track) (2023)."},{"key":"2026092110251413000_2026.09.01.748580v2.52","doi-asserted-by":"crossref","unstructured":"Abowd, J. M. The U.S. Census Bureau adopts differential privacy. Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining (2018).","DOI":"10.1145\/3219819.3226070"},{"key":"2026092110251413000_2026.09.01.748580v2.53","unstructured":"Jordon, J. et al. Hide-and-seek privacy challenge: synthetic data generation vs. patient re-identification. in Proceedings of the NeurIPS 2020 Competition and Demonstration Track Vol. 133 206\u2013215 (PMLR, 2021)."},{"key":"2026092110251413000_2026.09.01.748580v2.54","unstructured":"Stadler, T. , Oprisanu, B. & Troncoso, C. Synthetic data \u2013 anonymisation groundhog day. in 31st USENIX Security Symposium (USENIX Security 22) (USENIX Association, 2022)."},{"key":"2026092110251413000_2026.09.01.748580v2.55","doi-asserted-by":"publisher","DOI":"10.1080\/01621459.1963.10500845"},{"key":"2026092110251413000_2026.09.01.748580v2.56","doi-asserted-by":"publisher","DOI":"10.1137\/090771806"},{"key":"2026092110251413000_2026.09.01.748580v2.57","doi-asserted-by":"publisher","DOI":"10.1016\/j.jamda.2022.03.018"},{"key":"2026092110251413000_2026.09.01.748580v2.58","doi-asserted-by":"publisher","DOI":"10.1515\/bmte.1995.40.s1.317"},{"key":"2026092110251413000_2026.09.01.748580v2.59","doi-asserted-by":"publisher","DOI":"10.1038\/s44360-026-00096-z"}],"container-title":[],"original-title":[],"link":[{"URL":"https:\/\/syndication.highwire.org\/content\/doi\/10.64898\/2026.09.01.748580","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,9,21]],"date-time":"2026-09-21T17:25:29Z","timestamp":1790011529000},"score":1,"resource":{"primary":{"URL":"http:\/\/biorxiv.org\/lookup\/doi\/10.64898\/2026.09.01.748580"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,9,5]]},"references-count":59,"URL":"https:\/\/doi.org\/10.64898\/2026.09.01.748580","relation":{},"subject":[],"published":{"date-parts":[[2026,9,5]]},"subtype":"preprint"}}