SemiAnalysis identifies matching public safety results for 31 of 857 AI model releases in an assessment published on October 8, 2026. Its count covers nine Chinese developers, yielding a disclosure rate of 3.6%.
The authors examine releases from 2021 through September 15, 2026. They count substantive results linked to a specific model, rather than general statements about safety training.
According to their assessment, nine result reports were available by release time. CNA also reports the findings, citing Reuters.
This measures disclosure, not every internal test. Different model-variant naming limits company comparisons. AI Incidents has not independently replicated the count, and absent publication does not establish loss of control.