[DigitalToday reporter Chi-gyu Hwang] One of the most important yet hardest-to-grasp issues in AI security is how quickly China is developing top-tier AI models, Axios reported on June 23 (local time).

Z.ai, a Chinese company, recently released an open-source model called GLM-5.2 that drew attention for showing agentic capabilities comparable to Anthropic Opus4.8. It also drew praise in Silicon Valley.

But questions remain over how quickly China will narrow the gap with the United States.

According to the Axios report, the biggest disagreement among experts is not whether China is catching up to the United States, but how fast it is doing so.

Stanford’s AI Index report assessed that Chinese models have rapidly closed the quality gap over the past year, leaving the U.S. advantage nearly gone.

David Sacks, former White House AI czar, said in June the U.S. lead is only 6 to 9 months.

Another view holds that benchmark results alone do not show China has solved the compute resources and infrastructure issues needed for frontier competition.

Alex Stamos, former Facebook chief security officer, told Axios that it is quite possible China has truly outstanding models it has not disclosed publicly. "Thinking we're the best just because we're Americans is arrogant and foolish," he said.

Pukar Hamal, founder and CEO of SecurityPal AI, sees it differently. He said China lacks the cutting-edge chips and vast data needed to develop competitive frontier models. "So far, U.S. companies are the ones with access to the most chips and data," he said.

The Trump administration is still holding internal discussions over how to release Anthropic Fable5 and Mythos5, citing safety and national security concerns. OpenAI recently updated its cyber model GPT-5.5-Cyber in a direction that allows companies to make broader use of it.