If the keywords of global AI governance from 2023 to 2025 were "initiatives", "principles" and "frameworks", the keywords of 2026 have become "in force", "enforcement" and "fines". As China's AI content labeling measures take effect and the EU AI Act enters phased application, AI governance is shifting from soft consensus to hard constraints.
A realistic question for enterprises: the old rhythm of "launch first, comply later" may directly cross regulatory red lines after 2026.
China: Mandatory content labeling
On September 1, 2025, the Measures for Labeling AI-Generated Synthetic Content jointly issued by four departments took effect, together with a compulsory national standard. This means:
- Explicit labeling: AI-generated text, images, audio and video must be marked with user-perceptible watermarks or captions;
- Implicit labeling: generation metadata must be embedded in files for platform verification and tracing;
- Platform obligations: distribution platforms must establish detection and labeling mechanisms.
Meanwhile, the State Council's "AI+" Action Opinions set a target of over 70% adoption of new-generation intelligent agents and terminals by 2027 — dual-wheel drive of development and governance defines the Chinese approach, covering the full chain of training, service and distribution.
EU: High-risk countdown
The EU AI Act, effective August 2024, applies by risk tier: prohibitions from February 2025, obligations for general-purpose models from August 2, 2025, and — critically — most obligations for high-risk AI systems become fully applicable on August 2, 2026, covering education, recruitment, credit scoring and critical infrastructure, with fines up to 7% of global turnover.
US: Federal swing, state patchwork
Unlike China and the EU, US federal policy keeps swinging with administrations, while substantive rules land at state level — Colorado's AI Act, California's transparency laws — creating a fifty-state compliance patchwork for global enterprises. On coordination, China proposed the Global AI Governance Initiative, the UN adopted AI capacity-building resolutions, and 61 countries signed the declaration at the 2025 Paris AI Action Summit.
Enterprise compliance checklist
📋 Six self-check items
· Explicit + implicit labeling capability for generated content?
· Filing and security assessment completed for generative AI services?
· Complete records of training data sources and authorization?
· High-risk classification screening started for EU business?
· Explainable, auditable and traceable AI decision chains?
· Emergency response mechanism for AI security incidents?
Governance is a passport, not a brake
NineZenith observes that compliance capability has become a hard criterion in government and enterprise AI procurement. Zenith-Safety integrates content labeling, audit logs and security assessment support, combined with Zenith-Act's private deployment and full-chain tracing — governance and innovation are two sides of the same coin.