China’s Z.ai says its GLM matches Mythos

GLM-5.2 open-weight – China’s Zhipu AI says its open-weight GLM-5.2 can match Anthropic’s Mythos in some cybersecurity bug-finding tests, even as it falls behind in broader tasks. US officials have been trying to restrict China’s access to advanced AI models and the hardware needed

China’s Zhipu AI, operating under the name Z.ai, has released an open-weight model called GLM-5.2 — and with it, a claim that has the US intelligence and defense world watching closely.

Researchers supporting Z.ai’s message say GLM-5.2 can match Anthropic’s Mythos in certain bug-finding and cybersecurity scenarios. The comparison isn’t framed as a total catch-up. The model is described as lagging behind Anthropic’s and OpenAI’s systems in more general tasks. Still. the thrust is clear: China says it has narrowed the capabilities gap in the kind of work that can expose weaknesses in software and systems.

That “narrowing” is exactly what makes the development land like a warning bell in Washington. The US government has worked to restrict China’s access to powerful AI models such as Anthropic’s Mythos and Fable. along with the hardware needed to train and run them. Within that posture, the Trump administration has treated advanced AI models that can identify vulnerabilities as serious national security threats.

The model’s release method is part of what worries officials. GLM is an open-weight model, which means it can be downloaded and run by anyone on readily available hardware. That flexibility can bring benefits for researchers and power users who want deep access. It also strips away the kind of controlled environment that can limit how powerful models are tested and deployed.

The result, as critics see it, is a pathway for abuse. Because GLM can be run with little oversight, it can be “ripe for abuse” by bad actors — people who don’t need permission to experiment, fine-tune, or probe.

Even outside China, the same security pressure is showing up in the way US-linked institutions view advanced models. Recently, OpenAI unveiled GPT-5.6, and that announcement also raised concerns about potential misuse and led to limited access to the system.

GLM-5.2 now sits inside that broader tension: more capability, more accessibility, and — to the US government — more risk. Z.ai’s claim that its model can keep pace with Mythos in targeted cybersecurity scenarios may be measured in benchmarks. In Washington’s framing, the stakes are measured in what someone can do with it once it’s within reach.

Zhipu AI Z.ai GLM-5.2 Mythos cybersecurity open-weight model national security Anthropic Fable Trump administration hardware restrictions GPT-5.6 misuse risk