In December 2025, China's National Healthcare Security Administration issued guidance that formally classifies "AI-assisted diagnosis" as an extended item within pathological diagnosis and folds it into the pricing framework for pathological diagnostic services. The move signals that regulators are treating clinical AI as a billable, standardised service rather than an experimental add-on, pushing healthcare institutions and companies to clarify where the technology fits inside existing medical and legal structures.
Policy shifts and market focus
Alan Zhou, head of the life sciences and healthcare practice at Global Law Office, said healthcare was one of the first verticals where AI achieved real-world deployment, and it is now deepening along a clear trajectory under policy guidance. He described the progression as moving "from pilot demonstrations at large medical institutions, expanding to grassroots empowerment and community-level penetration; from 'light healthcare' scenarios such as health consultation, advancing toward the closed-loop integration of medical service systems; from solo exploration, evolving into collaborative governance among enterprises, healthcare institutions, and government."
Zhou identifies four primary areas where domestic AI for Healthcare companies are concentrating their products: clinical decision support, medical imaging, health management and patient-facing services, and drug discovery. These directions target improved diagnostic efficiency and optimised health management while steering clear of higher-risk segments of medical practice - offering what he calls a relatively prudent path to deployment.
Data compliance: the consent dilemma
The sensitivity of medical data makes compliance a baseline requirement for survival. Zhou pointed to data sourcing and its legality as a core difficulty. "The development and deployment of AI technology requires the use of vast quantities of data for training, and data in the medical and pharmaceutical fields frequently involves the sensitive personal information of patients and clinical trial participants," he said. "As data usage demands grow rapidly alongside technological advancement, regulatory enforcement centred on informed consent under the Personal Information Protection Law is simultaneously intensifying."
Faced with this tension, companies are exploring multiple compliance pathways. Beyond fulfilling Personal Information Protection Law requirements, Zhou noted that companies "in practice may lean toward using publicly available database resources, or establishing cooperative relationships with healthcare institutions." He added that resolving the data dilemma requires support and guidance from regulators to facilitate the development and sharing of high-quality datasets, releasing the value locked inside the data.
Regulatory boundaries for AI services
On the technology side, algorithmic and large-model compliance has become a regulatory priority. Rules including the Provisions on the Administration of Algorithmic Recommendation for Internet Information Services impose requirements around transparency and fairness. The Interim Measures for the Ethical Review and Service of AI Technology, which took effect in April this year, further establish mandatory upfront ethics review requirements.
Zhou highlighted two major compliance issues companies should keep in focus. First, unlicensed medical practice and prescription management. Under the Detailed Rules for the Supervision of Internet Diagnosis and Treatment, effective February 2022, AI software may not impersonate or substitute for a physician in providing medical services, nor may it use AI to auto-generate prescriptions. "Most services currently available in the market still fall within the categories of health management and consultation, thereby avoiding risks associated with unlicensed medical practice and prescription management," he said.
Second, pharmaceutical and medical device advertising and promotion. Zhou cautioned that even when content does not constitute an advertisement, companies remain exposed to unfair competition risks from false or misleading claims. Given the characteristics of generative AI models, he said companies must "exercise quality control over training data to ensure that data sources are lawful, content is accurate, and nothing is misleading," and "carefully design and continuously monitor algorithmic mechanisms to prevent models from automatically generating product marketing content."
Anti-corruption and anti-bribery remain perennially relevant because of the distinctive interaction framework among companies, institutions, and professionals in the pharmaceutical sector. Zhou cited as a high-risk scenario the practice of "inviting physicians to collaborate on AI application research and development in a particular disease area, and using that engagement to pay remuneration exceeding fair market value."
Why this matters for healthcare professionals
The pricing framework for AI-assisted diagnosis will directly affect how pathology departments budget for and deploy these tools, shifting them from pilot projects to operational line items. At the same time, the strict boundaries around unlicensed practice and prescription generation mean that clinicians, not algorithms, remain the legal decision-makers - protecting professional accountability but also requiring that physicians understand exactly where the AI's output ends and their own judgment begins. For anyone involved in procuring or using clinical AI, the compliance burden around data consent, advertising claims, and fair-value collaboration with industry partners is not a background concern; it is the condition on which market access rests.
Your membership also unlocks:
