Just weeks after the U.S. government restricted access to Anthropic’s latest AI models due to national security concerns, China unveiled a comparable open-weight artificial intelligence system that rivals—and in some cases surpasses—American counterparts in coding and cybersecurity capabilities. The model, **GLM-5.2**, was released by Beijing-based AI firm **Z.ai** and is designed for long-horizon software engineering and agentic coding tasks [1][3].
GLM-5.2 is distributed as an **open-weight model**, meaning users can download its weights, run it on their own servers, and fine-tune or modify it for specific purposes [2]. Unlike many proprietary models, it is licensed under **MIT terms**, allowing unrestricted global use without regional limitations [5]. This openness challenges the assumption that advanced AI control can be confined within national borders, as the model enables decentralized deployment and customization [6].
Security testing by **Semgrep** revealed that GLM-5.2 achieved a **39% performance score** in vulnerability detection, outperforming Anthropic’s **Claude Code** at 32% [6]. Another firm, **Graphistri**, reported that GLM-5.2 matched the problem-solving rate of Anthropic’s **Opus** series in cybersecurity investigations, while costing **less than half** the price [6]. However, Semgrep cautioned that these results were based on a single dataset under specific conditions and do not necessarily mean GLM-5.2 is superior across all security tasks [6].
The U.S. export restrictions on Anthropic’s **Fable 5** and **Mitos 5**, enacted on June 12, 2026, were justified under national security authority to prevent foreign access to powerful AI systems [6]. In response, Anthropic halted access to both models entirely, citing its inability to verify user nationality in real time [6]. The rapid emergence of GLM-5.2 underscores how open-weight AI models can circumvent such controls, raising new challenges for global AI governance and cybersecurity policy
