Study of 857 releases from nine Chinese AI labs (2021‑September 2026) finds only 3.6% included developer safety data and just 1.1% at launch (SemiAnalysis)
ai-safety
| Source: Techmeme | Original article
A review of 857 AI releases from nine Chinese labs between 2021 and Sep 2026 found only 3.6% included developer safety results and just 1.1% did so at launch.
A new study by California‑based research firm SemiAnalysis reveals that safety documentation is strikingly rare among Chinese generative‑AI releases. The analysis covered 857 model launches from nine leading Chinese AI companies—Alibaba, ByteDance, Tencent, Baidu, DeepSeek, Moonshot AI, Z.AI, MiniMax and StepFun—spanning 2021 to 15 September 2026. Only 31 releases (3.6 %) included any published safety evaluation, and a mere nine (1.1 %) provided safety results at the moment of launch.
The findings arrive as the global AI community grapples with mounting safety concerns. Earlier this month, Dario Amodei’s essay warned that frontier labs must treat safety as a deliberate, non‑negotiable priority. In the same period, OpenAI dismissed three safety researchers over internal disputes, and Anthropic began offering free security scans for open‑source projects. The Chinese context is further underscored by the recent decision of the ARTEX developer to close‑source its agent after CrowdStrike linked it to attacks on South Korean banks. Together, these events highlight a widening gap between rapid model deployment and the transparency needed to assess risks.
The SemiAnalysis report suggests that Chinese firms are lagging behind emerging best‑practice norms, potentially exposing users and downstream applications to unchecked hazards. Regulators in China have hinted at tighter AI oversight, and the data may prompt policymakers to require pre‑launch safety disclosures. Industry observers will watch for any formal guidance from the Ministry of Industry and Information Technology, as well as possible voluntary shifts by the labs themselves.
If the trend continues, the disparity between model proliferation and safety reporting could intensify calls for international standards, echoing the broader debate sparked by recent safety‑related controversies in the United States and Europe. The next few months should reveal whether Chinese AI developers adjust their release practices or face regulatory pressure to close the transparency gap.
Sources
Back to AIPULSEN