Research · Updated 11 Oct, 08:50 am IST
Report: Chinese AI firms published safety tests for just 3.6% of models
Why it matters for readers: For general readers, it highlights that many AI models are released without publicly linked safety test results, raising oversight questions.
- SemiAnalysis reviewed 857 models released between 2021 and September 15 and across nine Chinese AI companies.1
- Only 31 releases (3.6%) had a published safety-evaluation result that could be matched to a specific model.1
- Just nine releases (1.1%) had safety-evaluation results available at or before launch.1
- Researchers found no public safety disclosure tied to a named model for 813 of the releases, though companies may have tested privately.1
- SemiAnalysis defined disclosures as specific, model‑named test results for harmful output, jailbreak resistance, toxicity, privacy, refusal behaviour or dangerous capabilities, excluding general claims of safety training.1
Get a brief like this every morning
Uzha reads hundreds of sources and gives you the stories that matter for your work, with every source linked. Free.
Get started