The narrowing capability gap comes as Chinese AI adoption accelerates, driven in part by businesses looking for lower-cost alternatives. Companies, including Microsoft, are evaluating how Chinese AI models could be offered through their platforms, a shift that could reshape competition across the AI industry.
“China is making sure that the gap becomes smaller and smaller over time,” said Lior Div, chief executive officer of cybersecurity company 7AI.
The progress has heightened attention on AI’s role in software security. Researchers say increasingly capable models can help identify vulnerabilities before attackers exploit them, but the same capabilities also raise concerns that cybercriminals could use the technology to discover software flaws more efficiently.
Unlike models from Anthropic or OpenAI, GLM-5.2 is an open-weight model, allowing organizations to download, modify and operate it on their own hardware without relying on a hosted service. While that flexibility appeals to enterprises seeking greater control, security experts say it also makes the technology easier for malicious actors to deploy privately.
According to OpenRouter, GLM-5.2 has become one of the platform’s 10 most-used AI models. Cybersecurity company Semgrep said the model outperformed Anthropic’s Claude Opus 4.8 in some benchmark tests, and researchers found that with additional prompting, both Opus 4.8 and GLM-5.2 could match Mythos in bug-finding performance.
China’s cybersecurity sector is also advancing its own AI security tools. This week, 360 Security Technology introduced a bug-finding system called Tulongfeng, which the company said delivers performance comparable to Mythos.
“This kind of powerful weapon that can alter the landscape of cyberwarfare can’t remain solely in American hands,” 360 Security Chief Executive Zhou Hongyi said at a cybersecurity conference in Beijing. He added that China would face unacceptable risks if U.S. organizations could deploy advanced AI to analyze Chinese networks while Chinese companies lacked equivalent capabilities.
The developments come as U.S. AI companies face tighter government oversight. OpenAI recently said it is limiting access to GPT-5.6 during a preview period while a new government review process is implemented. Anthropic also restricted access to one of its latest general-purpose models after the Trump administration prohibited foreign entities from using it because of security concerns. The administration later restored access to the related Mythos 5 model for some trusted organizations.
Those restrictions have drawn criticism from some technology policy experts, who argue they could weaken the competitive position of U.S. AI companies while Chinese alternatives continue to gain traction.
“Banning Fable while selling chips China needs to develop its own version is a gift to China,” said Saif Khan, a distinguished technology fellow at the Institute for Progress who previously worked on export restrictions during the Biden administration. He said the United States should maximize deployment of Mythos and similar models to strengthen its cybersecurity defenses.
The administration has also signaled increased scrutiny of Chinese open-weight AI models. “Our administration is very much focused on Chinese open-source models,” said Jacob Helberg, undersecretary of state for economic affairs. “It’s something that we’re tracking very closely.”
Meanwhile, the Pentagon recently announced an agreement with Reflection AI, one of the few U.S. developers of open-weight AI models, for classified government use alongside similar partnerships.
Some AI researchers argue that restricting access to advanced U.S. cybersecurity models while Chinese open-weight alternatives remain widely available could influence adoption decisions across the industry.
“It is incentivizing companies across the globe to use cheaper but very capable Chinese open-weight models, while at the same time undermining the U.S. AI industry,” said Niels Provos, a researcher who previously led security teams at Google and Stripe. “I don’t understand it.”
This analysis is based on reporting from WSJ.
Image courtesy of Unsplash.
This article was generated with AI assistance and reviewed for accuracy and quality.
