Chinese AI Model Matches Mythos in Cybersecurity Bug-Finding

According to ZeroHedge, Chinese AI company Zhipu AI released an open-weight model that matches the performance of Anthropic's restricted Mythos 5 model in some cybersecurity scenarios, while Anthropic has been forced to shut down its latest general-use models for over two weeks after its de-tuned public-facing Fable 5 model could be jailbroken into its unrestricted Mythos 5 form to perform tasks that pose security risks. The company, backed by Alibaba and Tencent, released GLM-5.2, which can match the latest US models when it comes to finding security bugs, though it still lags at other tasks, according to the Wall Street Journal.

Key takeaways
Zhipu AI released GLM-5.2, an open-weight model that matches Mythos 5 in some cybersecurity bug-finding scenarios, according to the source context.
GLM-5.2 ranked as one of the 10 most-used AI models according to OpenRouter data, and in some benchmarking tests by Semgrep, it bested Anthropic's Claude Opus 4.8 model released in May.
The model is open-weight, meaning it can be downloaded, run on hardware by anyone, and modified without supervision, raising security considerations.
The Trump administration restored some access to Mythos 5 for trusted entities while OpenAI limited access to GPT 5.6 after security concerns were raised, according to the source context.

Table of Contents
What the source confirmed
Open-weight model access and security implications
US policy response and industry impact
Capability gap and competitive landscape
What to watch next

What the source confirmed

The source context states that Zhipu AI, also known as Z.ai, released GLM-5.2, an open-weight model that can match the latest US models when it comes to finding security bugs. According to cybersecurity company Semgrep, GLM-5.2 bested Anthropic's Claude Opus 4.8 model in some benchmarking tests. When given further instructions, Opus 4.8 and GLM-5.2 can match Mythos in bug-finding ability, according to researchers cited in the source context. The source notes that GLM-5.2 has ranked as one of the 10 most-used AI models according to data from OpenRouter, a company that provides access to more than 400 AI models.

The source context reports that Anthropic has been forced to shut down its latest general-use models for over two weeks after it emerged that the company's de-tuned public-facing Fable 5 model could be jailbroken into its unrestricted form, Mythos 5, to perform tasks that pose security risks. Goldman Sachs Delta One head Rich Privorotsky framed GLM-5.2 as a Chinese shock to the system, stating that the model appears highly competitive on SWE benchmarks relative to some of the latest private models, and that the gap between open and closed models continues to narrow, according to the source context.

Open-weight model access and security implications

The source context emphasizes that GLM-5.2 is an open-weight model, meaning it can be downloaded and run on hardware by anyone and can be modified and used without supervision. The source notes that hackers are undoubtedly loving this capability. Zhou Hongyi, CEO of Chinese cybersecurity company 360 Security Technology, stated at a cybersecurity conference in Beijing that this kind of powerful weapon that can alter the landscape of cyberwarfare cannot remain solely in American hands, according to the source context. His company released a new bug-finding tool called Tulongfeng, which it claims is comparable to Mythos when it comes to finding bugs.

US policy response and industry impact

The source context reports that the Trump administration restored some access to Mythos 5 for trusted entities after initially restricting it. Among the Mythos 5 and Fable 5 users that had lost access before Friday's decision to restore Mythos 5 access for some trusted entities was the National Security Agency, which had been testing the tools and found them impressive in trials, according to people familiar with the matter cited in the source context. OpenAI on Friday said it will now limit access to its latest model, GPT 5.6, after Trump administration officials raised security concerns. The company warned that the government's current case-by-case evaluation process is not a good long-term solution, but they are adhering to it following a recent executive order focused on security and model oversight, according to the source context.

Saif Khan, a distinguished technology fellow at the Institute for Progress think tank who focused on export restrictions under the Biden administration, stated that banning Fable while selling chips China needs to develop its own version is a gift to China, according to the source context. Khan says that the US needs to maximize the use of Mythos and similar models to harden cyber defenses while it can. Critics of the White House approach have said it has been lax in restricting use of Chinese open-weight models from companies such as DeepSeek and Zhipu, which are popular among US businesses, according to the source context. The source notes that a host of companies, including Microsoft, are weighing how they can offer Chinese models on their platforms, a development that is set to alter the balance of power among tech companies.

Capability gap and competitive landscape

The source context states that overall, the capability gap between top US models and those built by Chinese companies has narrowed significantly, and use of Chinese AI systems has surged as businesses seek to rein in runaway costs. Artificial Analysis's new knowledge work benchmark rated GLM-5.2 higher than GPT 5.5, according to the source context. The source notes that GLM-5.2 is not quite cutting edge, but the gap between open and closed models continues to narrow, and it represents a big leap in capability and a clear sign the field is narrowing.

For readers following broader market updates , the competitive dynamics between US and Chinese AI companies can influence how technology investors evaluate execution risk, cost structure, and strategic positioning across the AI infrastructure and application layers. The source context does not provide specific revenue impact, market share data, or stock price reactions for the companies mentioned. However, the shift toward open-weight models and the narrowing capability gap represent structural changes in how AI capabilities are distributed and accessed, which can matter for long-term investment analysis and technology adoption patterns across enterprise and government sectors.

What to watch next

Market readers and technology investors may watch for future disclosures from Anthropic, OpenAI, Microsoft, and other US AI companies regarding their model access policies, security protocols, and responses to the evolving regulatory environment. The source context indicates that the Trump administration's case-by-case evaluation process is not viewed as a long-term solution by industry participants, suggesting that further policy clarification or regulatory framework development may be forthcoming. Readers may also monitor benchmarking data from cybersecurity firms such as Semgrep and model usage statistics from platforms such as OpenRouter to track the adoption and performance of open-weight models relative to proprietary alternatives.

Additional areas to watch include any future statements from the National Security Agency or other government entities regarding their use of advanced AI models for cybersecurity purposes, as well as any announcements from Microsoft or other platform providers regarding their plans to offer Chinese AI models. The source context does not provide a timeline for these potential developments, but the rapid pace of model releases and policy responses suggests that the landscape may continue to evolve quickly. Investors and cybersecurity professionals may also watch for any updates on the security status of Anthropic's Fable 5 and Mythos 5 models, as well as any new bug-finding tools or capabilities released by Chinese or US companies in response to the competitive dynamics described in the source context.

Read original source