One of the most consequential unknowns in AI security is how quickly China will close the gap to frontier AI models.
Why it matters
Five Eyes leaders warned that the world could be only months away from AI models that substantially accelerate cyber threats. Preparations, however, are being slowed by infighting in Washington and industry-wide confusion about how to measure AI risk.
Recent developments
A new Chinese open-source model, GLM-5.2, drew major attention over the weekend for reportedly matching agentic capabilities seen in models such as Anthropic's Opus 4.8. The release earned praise from some Silicon Valley figures and renewed questions about the speed of China's catch-up.
At the same time, the Trump administration remains in debate over the conditions for releasing Anthropic's Fable 5 and Mythos 5, citing safety and national security concerns.
The big picture
The central dispute is not whether China is catching up, but how fast it is doing so. Stanford's AI Index Report indicates that Chinese models have rapidly improved in quality over the past year and have in many cases erased the U.S. advantage. Former White House AI czar David Sacks said this month that the U.S. currently holds roughly a six- to nine-month lead over China.
Others caution that benchmark gains alone do not demonstrate that China has overcome the compute and infrastructure challenges required to compete at the true frontier.
Expert observations
Alex Stamos, former Facebook security chief, told Axios that it is possible China has privately developed systems that are very strong, and it would be arrogant to assume the best technology exists only because it is American. He added that Chinese military hackers are likely "laughing" at internal U.S. disputes and reciprocal access restrictions.
Risks and economic effects
Officials and lawmakers are concerned China could deploy powerful AI to boost surveillance, cyber operations, and military decision-making. China's embrace of open-source models could also make its AI ecosystem more economically attractive globally, particularly for companies seeking alternatives to costly U.S. frontier models.
The surprise risk
Much of the debate centers on the risk of being surprised by rapid improvements rather than on current capabilities. An open-source security researcher, speaking anonymously because he was not authorized to speak for his employer, told Axios he uses frontier models to track how persuasion, social engineering, and vulnerability discovery capabilities are evolving. He worries that restricting access to cutting-edge U.S. systems as Chinese models improve could leave defenders with fewer tools to anticipate what comes next.
Counterpoints
Not everyone sees China's progress as an imminent threat to U.S. AI leadership. Pukar Hamal, founder and CEO of SecurityPal AI, told Axios that China lacks the bleeding-edge chips and the massive datasets needed to build a competitive frontier model. He noted that, so far, American companies have access to the most chips and the most data.
Reality check
Researchers can already find many vulnerabilities without using advanced models, which are often costlier and harder to access. AI-powered security firm Aisle claimed last week that its agentic capabilities are outperforming Mythos in several tests.
What to watch
OpenAI recently made its cyber model, GPT-5.5-Cyber, more permissive and capable; on several key benchmarks it outperforms Anthropic's Mythos 5.
Conclusion
The acceleration of AI development and the rise of Chinese open-source models heighten pressure on U.S. policymakers and industry to clarify how to measure risk and manage access. The core question remains timing: how long do the United States and its allies have to adapt to a rapidly changing technological landscape?
