China's first medical AI data security standard, "Medical Artificial Intelligence Application: Data Security and Privacy Protection Specification," is recruiting participating units

Add crowdsourced testing promotional copy Add crowdsourced testing promotional copy end Add App-exclusive traffic guidance

(Source: AI Frontier Early Knowledge)article_adlist[

Source | Zhihe Standardization Construction ]article_adlist

Author | Zhihe Standards Center

Following the issuance this year by five national departments of the "Measures for the Administration of Data Security and Personal Information Protection in Medical and Health Institutions (Trial)," local implementation rules are accelerating their follow-up. On September 21, 2026, the Zhejiang Provincial Health Commission and two other departments jointly issued the "Measures for the Administration of External Data Cooperation by Medical and Health Institutions of Zhejiang Province (Trial)." For the first time, the document explicitly includes "model training" within the regulatory scope of external data cooperation by medical institutions: research and development must be carried out in a closed environment within the domain, intermediate and temporary training data must be protected according to the standards for raw data, and training without de-identification and the illicit transmission of training data abroad are strictly prohibited. Data compliance for medical AI is moving from principled requirements toward executable, inspectable detailed rules.

However, medical data must pass through multiple stages and many hands from collection, de-identification, and annotation to training and deployment. What records should be kept at each stage, who should confirm them, and to what extent they qualify as compliant currently all lack a unified basis; model reverse attacks may infer personal information in training data from outputs, and anonymized data may still be re-identified through linkage and splicing—to self-prove that these risks are "already under control," it is unrealistic to rely solely on medical institutions figuring things out on their own. The medical industry urgently needs a unified methodology for medical AI data security protection.

Against this backdrop, to promote the deep integration of "AI + healthcare," the nation's first data security and privacy protection standard focused on medical artificial intelligence applications, the "Data Security and Privacy Protection Specification for Medical Artificial Intelligence Applications," administered by the China Electronics Chamber of Commerce and organized and drafted by the Zhihe Standards Center, has emerged. Since its launch, the standard has attracted participation from multiple forces including medical institutions, medical devices, biopharmaceuticals, artificial intelligence technology, and security assurance, and has now gathered dozens of units to jointly participate in the standard drafting work.

Information and Application Security Protection Branch of the Chinese Health Information and Health Care Big Data Association

At present, the standard text has completed multiple rounds of revision and formed a draft for comments, and there is still a final opportunity to participate in the drafting.

This standard focuses on key links such as data classification and grading, full-lifecycle security management, application of privacy protection technologies, and definition of responsibilities in multi-party collaboration, building a security management system covering the entire process of data acquisition, annotation and preprocessing, training and fine-tuning, validation and evaluation, deployment and inference, continuous optimization, and exit and disposal, and provides implementation guidance for typical application scenarios such as medical imaging-assisted diagnosis, generative consultation, and wearable devices and remote monitoring, providing systematic data security and privacy protection specifications for medical artificial intelligence applications and promoting the standardized, regulated, and sustainable healthy development of medical artificial intelligence applications.

Core highlights of the standard

Building a three-dimensional governance architecture of "data regulatory category—data security protection level—application risk"

The standard establishes a classification and grading mechanism that synergistically links three dimensions: data regulatory category, data security protection level, and application risk level. It divides data security protection levels from low to high into levels one through four and establishes graded control mapping rules, so that data of different sensitivity and different risk scenarios can all find corresponding protection requirements, avoiding a one-size-fits-all compliance burden.

Clarifying full-lifecycle security management requirements

From project initiation, data introduction, annotation and preprocessing, training and fine-tuning, validation and evaluation, online deployment, operation and maintenance, to decommissioning and exit, the standard puts forward clear security management and technical requirements for every link of medical AI applications, helping medical institutions and AI enterprises build implementable and auditable full-process management mechanisms.

Defining the boundaries of responsibility in multi-party collaboration

The standard clearly defines the rights, obligations, and responsibility boundaries of personal health and medical data subjects, personal information processors, entrusted processors, and all parties in multi-party collaboration frameworks, and proposes compliance guidance for typical collaboration scenarios such as inter-hospital joint research and remote monitoring, effectively resolving the difficult issue of determining responsibility for "who is responsible when something goes wrong with the data."

Shoring up the security bottom line while unleashing data value

The standard establishes seven basic principles: legality and compliance, data security, privacy protection, risk-graded protection, controllability and auditability and traceable responsibility, ethical review and clinical safety, and balance between security protection and data usability. It coordinates the security bottom line with data usability, provides security and compliance support for the construction of high-quality medical datasets, and bridges the last mile of the secure circulation of medical data elements.

As the nation's first data security and privacy protection standard focused on medical artificial intelligence applications, the drafting work has received positive responses from core participants across the upstream, midstream, and downstream of the industry chain since its launch. More medical and health service institutions, medical device institutions, biopharmaceutical institutions, artificial intelligence and data technology service providers, security assurance and professional evaluation institutions, as well as units and experts concerned with medical artificial intelligence data security and privacy protection, are welcome to participate in jointly building the standard and collaboratively construct a clear and implementable medical AI data compliance system.article_adlist[

If you are interested in becoming a drafting unit/drafter of the "Standard"

Please scan the QR code to fill in the relevant information

END ]article_adlist Statement: Exclusive article of Sina.com, unauthorized reproduction is prohibited. tech_web_article_v2015_block_bottom Non-targeted 300*250 button 17/09 wenjing begin

Non-targeted 300*250 button end