Analysis of 857 releases from nine Chinese AI labs from 2021 to September 2026: just 3.6% included safety results from the developer and only 1.1% did at launch

Chinese AI labs' minimal disclosure of safety results means you cannot assess risks before deploying their models.

An analysis of 857 AI model releases from nine leading Chinese developers between 2021 and September 2026 reveals a significant gap in publicly available safety data. Only 3.6% (31 releases) included any form of published safety results from the developers, and a mere 1.1% (9 releases) had these results available at the time of launch. The remaining 94.9% of releases had no safety disclosure whatsoever. This trend holds across major hyperscalers like ByteDance and Alibaba, as well as startups such as DeepSeek and MiniMax, with no single entity showing a concentration of safety reporting. The pace of releases has dramatically increased, with a thirty-fold rise from Q1 2023 to Q3 2025, yet the number of releases with safety results has remained consistently low, never exceeding seven per quarter. Policy milestones in China have not demonstrably influenced the rate of safety disclosure for these frontier models.

Despite China's public pronouncements on AI safety and its new AI Safety Governance Framework 3.0 highlighting risks like self-improvement and behavioral deviations, the practical output from its leading AI labs shows a clear prioritization of rapid development over rigorous safety testing and disclosure. The analysis indicates that while China imposes strict regulations on AI applications and outputs for public-facing services, the frontier models are developed with minimal external safety validation. This suggests a bifurcated approach where public-facing AI is controlled, but the underlying advanced models lack transparency regarding their safety evaluations, creating a potential disconnect between regulatory rhetoric and actual development practices.

The stark contrast between China's official safety rhetoric and the empirical data on AI lab releases signals a global race where speed and capability advancement are paramount, potentially overshadowing safety concerns at the frontier. This lack of transparency means that international researchers, policymakers, and users have limited insight into the potential risks posed by China's most advanced AI systems, impacting trust and the ability to establish globally consistent safety standards. The findings challenge the assumption that stated safety commitments directly translate into development practices, suggesting a need for more direct verification mechanisms for AI safety claims, particularly from rapidly advancing national AI programs.