The Cyberspace Administration of China introduced the Artificial Intelligence Security Governance Framework 3.0 on September 14, 2026. This policy mandates that AI models must reject at least 95% of 2,000 sensitive prompts, including those related to discrimination and subversion, before public release.
The regulatory body has actively enforced compliance between April and September 2026. During this period, authorities removed over 5.61 million pieces of non-compliant AI-generated content, penalized more than 49,000 accounts, and took enforcement actions against over 2,400 websites and applications.
Additional regulations effective since July 2026 require human oversight for anthropomorphic AI services and restrict companion-style features for minors. Furthermore, all AI-generated content must carry explicit labels, a requirement that has been in place since September 2025.
The Chinese government classifies risks associated with artificial intelligence as potential national emergencies, equating them to natural disasters. This framework represents the third iteration of the country's AI governance structure, reflecting an increasingly granular and strict regulatory approach.
