Plain-English answer
The United States tends to govern healthcare AI by intended use, device status, claims, clinical risk, data practices, and deployment context. China tends to govern healthcare AI through a combination of medical-device registration, data and cybersecurity oversight, platform regulation, and ethics review for sensitive science and technology activities.
How the U.S. side works
FDA focuses on whether AI-enabled software is a device function and what evidence supports safety and effectiveness. The agency's 2025 draft guidance for AI-enabled device software functions emphasizes lifecycle information in marketing submissions, while FDA's transparency principles for machine-learning-enabled devices emphasize user understanding, human-AI team performance, and total product lifecycle risk management.
How the China side works
China's AI governance in healthcare sits inside a broader state governance model. NMPA medical-device rules matter when software is a regulated device, PIPL matters when health data is processed, and the national science-and-technology ethics review framework matters when AI work is ethically sensitive, socially significant, or safety-relevant.
Side-by-side comparison
| Dimension | United States | China | Strategic implication |
|---|---|---|---|
| Regulatory center | FDA device software, privacy, FTC claims, civil-rights, and institutional governance. | NMPA device registration, PIPL, data security, cybersecurity, platform rules, and ethics review. | Classify the AI use case before choosing evidence, privacy, and deployment controls. |
| Lifecycle issue | Model changes, monitoring, transparency, cybersecurity, and real-world performance. | Registration scope, data control, ethics review, cybersecurity, and state oversight. | AI governance must cover post-launch behavior, not just premarket clearance. |
| Trust question | Will clinicians, patients, payers, and regulators understand and rely on the AI safely? | Will authorities and hospitals accept the data, platform, ethics, and medical-device posture? | Evidence must be localized to the decision-maker and the clinical workflow. |
Current evidence and sources
- FDA - AI-enabled device software draft guidance: lifecycle management and marketing-submission recommendations for AI-enabled device software functions.
- FDA, Health Canada, and MHRA - transparency principles: principles for transparency and human-AI team performance in machine-learning-enabled medical devices.
- Gov.cn - China science and technology ethics review guideline: ethics review governance for sensitive science and technology activities.
Strategic meaning
Healthcare AI companies moving between the two countries need separate evidence, privacy, model-change, and deployment narratives. A model that is technically strong can still fail if the claim makes it a regulated device, if training data cannot be explained, if the workflow creates unsafe automation bias, or if cross-border data governance is unresolved.