Stories
China AI Developers Publish Safety Tests for Just 3.6 Percent of Model Releases
C
CNBC TV18•09-10-2026, 18:41
China AI Developers Publish Safety Tests for Just 3.6 Percent of Model Releases
- •A report by California-based research firm SemiAnalysis revealed that leading Chinese AI developers publicly disclosed model-specific safety-test results for only 3.6 percent of their releases.
- •Out of 857 models released between 2021 and September by nine major Chinese tech firms, only 31 had published safety-evaluation results tied to a specific model.
- •Just nine model releases had safety results available at or before their official launch, while 813 releases had no public safety disclosures.
- •The findings highlight growing international concerns regarding AI safety as security incidents involving autonomous AI agents intensify worldwide.
- •China's current AI Safety Governance Framework regulates applications and user effects but does not impose mandatory duties linked to frontier model capabilities.
Key Questions Answered
Leading Chinese AI developers have publicly disclosed model-specific safety-test results for only 3.6% of their model releases, according to a report by SemiAnalysis.
Out of 857 models reviewed by SemiAnalysis, just nine releases, or 1.1%, had safety-evaluation results available at or before their launch.
China’s AI Safety Governance Framework identifies risks such as models acquiring unauthorized system permissions or deceiving evaluators, but it does not impose mandatory duties linked to model capability.
