China AI developers publish safety tests for just 3.6% of model releases, report finds
A report from California-based technology research firm SemiAnalysis reveals that only a minuscule 3.6% of AI model releases by leading Chinese companies have publicly disclosed comprehensive safety tests. The study analyzed 857 models released between 2021 and September 15 by nine prominent Chinese AI firms - Alibaba, ByteDance, Tencent, Baidu, DeepSeek, Moonshot, Z.AI, MiniMax and StepFun.
Out of these, a mere 31 releases, equating to 1.1%, had safety-evaluation results published alongside the model's launch. The remaining 813 releases saw no such disclosures, despite the companies potentially conducting tests privately. The firm defined disclosures as specific results linked to a named model, including assessments of harmful output, jailbreak resistance, toxicity, privacy, refusal behavior or dangerous capabilities.
The study comes as autonomous AI agents, which operate with limited human intervention, have sparked global discussions about the need for slower development to ensure model safety. China's AI Safety Governance Framework acknowledges risks such as unauthorized system access, deception of evaluators, concealment of capabilities, and bypassing safety measures, but does not mandate model capability assessments.
The report highlights that no major Chinese developer has released a frontier text model with publicly available dangerous capability tests across cyber, biological and loss-of-control risks, unlike leading US companies like OpenAI, Anthropic and Google DeepMind.
Written by urgent.news from The Indian Express's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.