Chinese AI Labs Fall Behind in Safety Transparency

Chinese AI labs Z.AI and Alibaba are trailing American pioneers like OpenAI and Anthropic in terms of transparency and AI safety disclosures. However, global market demands may push these companies to improve.

Lack of Transparency Among Chinese Companies

While models from companies like OpenAI and Anthropic exhibit unexpected behaviors, Chinese firms ranging from Alibaba to Z.AI adopt a less transparent stance regarding AI safety.

Market-valued Z.AI quietly released its first AI safety framework, marking a first among major Chinese model developers.

Limited Voluntary Disclosures

While American companies regularly publish model evaluations, such voluntary disclosures are quite rare among firms in the People's Republic of China.

According to a July report by consulting firm Concordia AI, only half of the top ten model developers shared safety evaluation results over the past year.

Safety Tests and Undisclosed Details

Although Z.AI delayed the release of its flagship GLM-5.3 model by two weeks in August to conduct safety tests, it did not share the implemented fixes with the public.

Beyond risks related to piracy, the company failed to provide broader safety evaluations and avoided publishing detailed system cards like those from Anthropic.

Resource Constraints and Technical Barriers

One reason Chinese offerings lag in safety disclosures is that they are only just beginning to approach US-level frontier capabilities.

As AI firms in the People's Republic of China face chip and capital shortages, safety testing is both expensive and requires high computational power.

Global Market Regulations and Sanctions

As their advanced open-weight models gain traction in foreign markets, Chinese companies will be forced to match their American rivals in transparency and governance.

The European AI Act and measures in South Korea targeting high-impact AI systems will make it mandatory for these companies to adopt comprehensive safety measures.