Safety · 2026-10-09

Chinese AI developers rarely publish safety test results

According to a report, published safety tests cover only 3.6% of new AI models from Chinese developers.

✨ What the report covers

Reuters reports that Chinese artificial intelligence developers publish safety test results for only 3.6% of their new models. This figure comes from a report that the outlet references.

Safety tests check whether a model can give dangerous advice, create harmful content, or help with fraud. When a developer publishes such results, users and researchers can assess the risks themselves.

📌 Why it matters

If nothing is known about a model's safety, it is hard to tell how much it can be trusted. This applies both to ordinary users and to companies that build such models into their services.

The issue of transparency is now being discussed in many countries that are working on rules for AI. News about who publishes what helps people better understand the tools they use.

Sources: 🔗 Google News