China Unveils Global AI Ethics Framework with Tiered Risk Classification — From Domestic Regulation to International Rule-Making

On July 22, 2026, China released its International AI Ethics Governance Action Plan at the World Artificial Intelligence Conference (WAIC) in Shanghai, proposing a tiered risk-based governance framework that classifies AI agents and embodied AI as separate regulatory categories for the first time. The move signals China's strategic pivot from domestic AI compliance to shaping international AI governance standards.

Core of the Plan: Tiered Risk Framework

Jointly guided by the Ministry of Industry and Information Technology (MIIT) and other government bodies, the plan proposes that AI systems at different risk levels should be subject to proportionally calibrated oversight. The framework divides AI systems into three risk tiers:

• Unacceptable Risk — AI applications banned from deployment, such as social credit scoring and real-time biometric surveillance
• High Risk — AI systems involving critical infrastructure, employment, healthcare, requiring mandatory filing and continuous auditing
• Limited / Minimal Risk — Transparency obligations only, with no additional compliance requirements

This structure closely mirrors the EU AI Act's four-tier classification, but with a stronger emphasis on government-led filing and auditing mechanisms rather than third-party conformity assessment.

AI Agents and Embodied AI as Independent Regulatory Categories

The plan's most notable innovation is the explicit classification of AI agents and embodied AI as distinct regulatory categories. Previously, major global frameworks — including the EU AI Act and China's own AI Agent Implementation Opinions — treated agents as an application form of AI systems rather than a separate regulatory object.

China's new plan argues that AI agents' autonomous decision-making capabilities and embodied AI's physical interaction capabilities introduce risks beyond those of traditional AI systems: unpredictability of multi-step actions, cumulative risk across a chain of operations, and the potential for physical harm in the real world. These systems therefore require "differentiated oversight" — additional obligations layered on top of the general AI risk framework.

Strategic Intent: From Rule-Taker to Rule-Maker

The timing of the plan is deliberate. Just one week earlier (July 15), China's AI Agent Implementation Opinions took effect as the world's first dedicated regulatory framework with recall authority. The new plan extends regulatory ambition from domestic to international:

• Urging governments worldwide to adopt tiered AI ethics governance
• Proposing an international AI ethics standards coordination mechanism
• Advancing multilateral AI governance dialogue within the UN framework
• Advocating equal voice for developing countries in AI governance

MLex analysis describes this as "a systematic effort by Beijing to shape global AI rules." With the EU AI Act already established as the de facto Western standard and the US building its AI governance approach through export controls and technology alliances, China is constructing its own AI governance discourse — centered on "tiered ethics" and framed by "development rights" and "digital sovereignty."

Enterprise Implications

For enterprises operating in or doing business with China, this plan means:

• Agent deployment requires compliance upfront — AI agents are no longer treated as ordinary AI systems; they need dedicated compliance assessment and filing
• Cross-jurisdictional compliance pressure intensifies — enterprises simultaneously subject to EU AI Act, China's AI agent regulations, and US state AI laws face triple compliance burdens
• Global regulatory fragmentation accelerates — differences between Chinese, EU, and US AI governance frameworks are hardening, reducing prospects for global standards
• Governance infrastructure demand surges — enterprises need governance tools that can map to multiple regulatory frameworks simultaneously

International Response

International reaction to China's new plan has been divided. The EU AI Office welcomed "any international initiative advancing AI ethics," while emphasizing that "any global framework must be grounded in democratic values and fundamental rights." The US Commerce Department declined direct comment but reiterated "US leadership in AI governance."

Notably, on the same day (July 22), OpenAI and Anthropic jointly warned US policymakers about the risks of Chinese open-weight model proliferation. This temporal convergence — China launching a global governance plan while US AI labs warn about Chinese model risks — marks AI governance as the new frontier of great-power competition.