China tightens AI ethics rules, forcing startups to overhaul compliance
China's Ministry of Industry and Information Technology and nine other government departments issued new trial measures in May 2026 that make AI ethics review mandatory for companies-a shift that leaves startups exposed to significant compliance risks.
The measures establish formal procedures for ethical review, specify which entities bear responsibility, and create mechanisms for ongoing oversight. They apply to any institution or enterprise engaged in AI-related activities.
Grace Wang, a partner at Zhong Lun Law Firm, said the rules mark a fundamental change: "AI ethics is no longer an optional 'bonus point' for companies, but a mandatory legal compliance baseline."
Startups face immediate exposure
Zou Danli, a partner at Commerce & Finance Law Offices, flagged a specific risk for early-stage companies. Many startups may not fully understand their obligations and could proceed with AI work without conducting required ethical reviews, exposing them to administrative penalties.
The measures create designated service centres where companies can commission external ethical reviews, training, and consulting. This addresses a practical problem: smaller AI firms often lack specialized ethics staff.
"These arrangements reduce the operational burden and costs associated with compliance," Zou said.
Expert review becomes mandatory for high-risk work
Articles 21 to 25 of the measures establish a two-tier system. High-risk AI activities must first pass internal review, then be submitted to government authorities or local bodies for independent expert re-examination.
Wang said this creates three enforcement layers: elevated review standards, stricter ongoing oversight, and binding compliance obligations across the entire lifecycle of research, launch, and operation.
"Companies can no longer rely solely on internal reviews to complete their compliance loop," Wang said. "They have to accept independent evaluation from external experts, significantly raising the compliance threshold."
Five vulnerability points
Companies face the highest risk at five stages: organizational setup, prior review, high-risk procedures, dynamic management, and registration and filing.
AI work involving human dignity, life and health, public order, or environmental protection can be deemed non-compliant if conducted without prior ethical review or complete documentation.
What companies should do now
Wang advised establishing an ethics governance framework immediately, strengthening pre-launch review and risk assessment, implementing re-examination procedures for high-risk projects, introducing dynamic monitoring, and fulfilling registration obligations.
"Ethics compliance should be embedded throughout the entire lifecycle of AI development, testing, deployment and operation," she said.
For legal teams, understanding these requirements is essential. AI for Legal professionals covers compliance frameworks and regulatory obligations. Paralegals managing AI compliance should familiarize themselves with the procedural requirements outlined in these measures.
Your membership also unlocks:
