China accelerates AI model releases amid safety transparency concerns

Chinese technology firms are rapidly accelerating the release of artificial intelligence models, with sixteen new models introduced by ten major companies in September 2026 alone. Notable releases include DeepSeek’s V4.1-Flash, a multimodal model utilizing a Mixture of Experts architecture with 552 billion parameters, and Xiaomi’s MiMo-V2.6 series.

However, a report by the U.S.-based research firm SemiAnalysis highlights significant gaps in transparency regarding AI safety. After reviewing 857 models released by major Chinese developers—including Alibaba, Tencent, Baidu, and ByteDance—between 2021 and September 2026, the study found that only 3.6% disclosed specific safety test results. Nearly 95% of models provided no safety disclosures at all.

The report suggests that China’s regulatory approach, such as the “Artificial Intelligence Safety Governance Framework 3.0” released by the National Cybersecurity Standardization Technical Committee, prioritizes innovation and development over mandatory safety reviews for frontier models. This “speed-first” regime contrasts with international discussions regarding the risks associated with advanced AI capabilities.