China closes the gap with Anthropic in cybersecurity

China’s artificial intelligence industry has taken a major leap, raising concerns in Washington as the global AI race intensifies.

Researchers say Zhipu AI’s GLM-5.2 can detect software security flaws as effectively as Anthropic’s Mythos in some cybersecurity tests. It even outperformed Claude Opus 4.8 in certain benchmark tests.

Although Chinese models still lag behind the best US systems across general AI tasks, experts say the technology gap is closing rapidly.

Why the breakthrough matters
AI-powered bug detection helps companies identify and fix software weaknesses before hackers can exploit them. However, the same technology can also help cybercriminals identify vulnerabilities more quickly.

GLM-5.2 is an open-weight model, meaning anyone can download, modify, and run it without restrictions. That makes it cheaper and more flexible for businesses, but also easier for hackers to misuse. The model has already become one of the world’s 10 most widely used AI models, reflecting its growing adoption.

Pressure on US AI policy
China’s rapid progress comes as the US limits access to some advanced AI models over national security concerns.

Some industry experts believe that restricting AI model access while allowing semiconductor sales to China creates a policy contradiction, as those chips can support the development of competitive domestic AI models.

Critics also argue that these limits could drive businesses toward cheaper Chinese alternatives, while some companies are exploring adding Chinese AI models to their platforms.

Experts warn that if the US limits access to its own AI systems while China continues to advance, it risks losing its long-standing leadership in AI and cybersecurity.

Anthropic CEO Dario Amodei has also warned that powerful AI models can pose serious security risks if widely misused.

The debate has already influenced US policy. The Trump administration temporarily banned foreign users from accessing Anthropic’s Fable 5 model, later restoring limited access to its Mythos 5 for trusted organizations.

On the same day, OpenAI, a competitor to Anthropic, unveiled GPT-5.6, making it available only to a limited number of government-approved recipients.