A Chinese technology company has said its latest software programme can match the performance of Anthropic’s most powerful tool when it comes to finding hidden weaknesses in computer systems. The claim, if independently verified, would represent a significant narrowing of the gap between American and Chinese capabilities in one of the most strategically sensitive areas of modern computing.
The company making the claim is Zhipu AI, also known as Z.ai. It was founded in 2019 and is based in Beijing. Its latest programme is called GLM-5.2. According to a report published by the Wall Street Journal, security researchers found that GLM-5.2 can perform at the same level as Anthropic’s Mythos when the task is identifying software security flaws. Mythos is the model that Anthropic has kept under the tightest restrictions, citing concerns about what could happen if it fell into the wrong hands.
This matters because finding software vulnerabilities is not a trivial task. When a weakness is discovered in a piece of software, it can be used either to defend systems by patching the problem before attackers find it, or to carry out attacks by exploiting it before a fix is in place. A model capable of identifying those weaknesses automatically, at scale, holds enormous value for both defenders and attackers. Mythos had been regarded as the most capable tool available for this kind of work.
Researchers were careful to place the claim in context. GLM-5.2 still falls behind Anthropic and OpenAI’s models across a broader range of tasks. The performance gap has not closed everywhere. But in the specific area of finding security bugs, the distance appears to have shrunk considerably. Some evaluations found the Chinese model performing above Anthropic’s Claude Opus 4.8 in certain security tests. With additional prompting, researchers said it could reach performance levels comparable to Mythos itself.
GLM-5.2 also has a characteristic that Mythos does not. It is open-source. Anyone can download it, modify it and run it on their own hardware without paying for access to a cloud service. That openness makes it accessible to businesses that cannot or will not pay for American tools. It also raises concerns that people with harmful intentions could adapt it for offensive purposes without any of the safeguards that Anthropic has built around its own restricted model.
A separate development at the same time underlined the competitive pressure China is applying in this space. The founder of Chinese cybersecurity company 360 Security Technology, Zhou Hongyi, appeared at the ISC.AI 2026 conference in Beijing and unveiled two new security tools under the brand name Yitian Tulong, a reference to a classic Chinese martial arts story. One tool, called Tulongfeng, was described by Zhou as China’s version of Mythos. He said it had been designed to find software vulnerabilities automatically.
A second tool was built to handle defensive operations and respond to incidents. Zhou did not describe Tulongfeng as having matched Mythos in all respects. Instead, he said his company had taken a different route, combining a language processing model with existing security expertise, vulnerability records and automated processes to achieve comparable results in the specific area of vulnerability discovery. He acknowledged that Chinese models still have a gap of between 20 and 30 per cent compared to American counterparts in overall capability. But he argued China could not afford to wait for that gap to close before developing its own security tools. “This kind of powerful weapon that can alter the landscape of cyberwarfare cannot remain solely in American hands,” he told the conference.
According to the report, the timing of these announcements carries particular weight given recent events. The American government imposed restrictions on Anthropic’s models in mid-June, cutting off global access to both Mythos and its consumer-facing model. While the United States government has since partially restored access to Mythos for a small group of vetted American organisations, the wider restrictions remain in place internationally. That situation has created an opening for Chinese alternatives.
See Also: Instagram new algorithm controls for feeds and reels give users new ways to shape what they see
Companies around the world that had been using or evaluating Anthropic’s tools suddenly found themselves without access. Some began looking at lower-cost alternatives, including Chinese models. The Wall Street Journal reported that companies including Microsoft are now considering how they might offer Chinese models on their own platforms. For American companies, this is an uncomfortable outcome. The restrictions intended to keep powerful tools out of Chinese hands may simultaneously be pushing international customers toward Chinese alternatives.
A researcher who previously led security teams at Google and Stripe made the point directly to the Wall Street Journal. Restricting access to American models, he said, is pushing companies across the world toward cheaper but increasingly capable Chinese open-source alternatives. “I don’t understand it,” he said.
The broader picture is one of a technology competition that has accelerated faster than most governments anticipated. American restrictions assumed that cutting off access to advanced chips would slow Chinese development significantly and preserve a meaningful lead for the United States. The emergence of capable Chinese models built within those constraints suggests the gap is closing on a shorter timeline than expected.
About the Author
Marcel Chidozie is a tech analyst and writer covering foreign news, fintech, and emerging technologies at TechRegard. Based in Nigeria, He's passionate about translating complex tech developments into compelling, accessible stories for diverse audiences. His work focuses on how technology shapes innovation across Africa and globally.
