Chinese AI Models Achieve Parity with US in Cybersecurity Capabilities, Report Finds

Chinese artificial intelligence models have reportedly reached parity with leading U.S. systems in cybersecurity capabilities, a development that could intensify pressure on the White House as it formulates domestic AI policy. Security researchers indicate that a recently released model from China's Zhipu AI, also known as Z.ai, demonstrates performance comparable to Anthropic’s flagship Mythos model in certain bug-finding scenarios.

While the Zhipu AI model, GLM-5.2, still lags behind U.S. industry giants like Anthropic and OpenAI in other areas, the overall performance gap has significantly diminished, according to The Wall Street Journal. Concurrently, a substantial influx of powerful yet accessible Chinese AI models is rapidly gaining traction among customers in the United States. Industry observers note that major tech companies, including Microsoft, are considering the integration of these Chinese systems onto their platforms, a move that could potentially alter the competitive dynamics within the technology sector.

According to OpenRouter, a platform offering access to over 400 AI models, GLM-5.2 ranks among the top 10 most utilized AI systems. Cybersecurity firm Semgrep reported that GLM-5.2 outperformed Anthropic’s Claude Opus 4.8 in specific benchmark tests. Further analysis by researchers suggests that with additional prompting, both Opus 4.8 and GLM-5.2 can achieve performance levels similar to Mythos in identifying software vulnerabilities.

On Wednesday, Chinese cybersecurity firm 360 Security Technology introduced Tulongfeng, a new bug-finding tool it claims performs on par with Mythos. These advancements have elicited concerns from national security officials and corporate executives.

"China is making sure that the gap becomes smaller and smaller over time," Lior Div, chief executive of cybersecurity company 7AI, told The Wall Street Journal. Guillermo Rauch, CEO of U.S.-based AI firm Vercel, expressed his surprise on X earlier this month, stating, "Genuinely impressed, almost shocked, at how good GLM 5.2 by @zai_org is at coding. This changes things."

The escalating capacity of AI to detect software flaws has amplified the urgency to leverage this technology for patching security weaknesses before malicious actors can exploit them. Researchers have cautioned that failure to do so could precipitate a scenario some have termed "bugmageddon."

Zhipu’s GLM-5.2 is an open-weight model, allowing for its download, operation, and modification on personal hardware without external oversight. This stands in contrast to models developed by Anthropic or OpenAI, which have more restricted access. Open-weight models offer greater control to organizations but also provide potential access to powerful tools for malicious actors.

"This kind of powerful weapon that can alter the landscape of cyberwarfare can’t remain solely in American hands," stated Zhou Hongyi, CEO of 360 Security, at a cybersecurity conference in Beijing, as reported by The Journal. Zhou argued that China would face unacceptable risks if U.S. organizations could scan critical Chinese networks using advanced AI models while Chinese companies lacked equivalent capabilities.

China's progress in AI development occurs as the U.S. government has implemented restrictions on the release of advanced AI models. Last Friday, OpenAI announced it was limiting access to its newest model, GPT-5.6, citing security concerns voiced by administration officials. The company described its current case-by-case review process as a temporary measure pending the implementation of an executive order on AI security and model oversight.

Similarly, one of Anthropic’s recent general-purpose models remained inaccessible for over two weeks following a ruling by the Trump administration that prohibited its use by any foreign entity or individual due to security risks. Anthropic ceased access to comply with this directive. On Friday, the administration reinstated limited access to a related Anthropic model, Mythos 5, for select users.

Critics have contended that the administration's actions toward a prominent U.S. AI company are counterproductive, particularly given the continued export of AI chips to China, which facilitates the country's rapid AI advancements. "Banning Fable while selling chips China needs to develop its own version is a gift to China," remarked Saif Khan, a technology fellow at the Institute for Progress who previously worked on export restrictions during the Biden administration. Khan suggested that the U.S. should maximize the utilization of Mythos and comparable models to bolster its cyber defenses while it holds a strategic advantage.

Critics of the White House's approach have also asserted that insufficient measures have been taken to limit the use of Chinese open-weight models from companies like DeepSeek and Zhipu, which have gained popularity among U.S. businesses. In a related development, the Pentagon recently announced an agreement with Reflection AI for classified applications, alongside several similar ventures, signaling a potential shift toward supporting domestic open-weight AI developers.

Meanwhile, AI users have expressed concerns that U.S. efforts to restrict access to increasingly sophisticated cybersecurity models could lead to the unavailability of critical AI tools. "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 formerly leading security teams at Google and Stripe. "I don’t understand it."