A Chinese AI model has attracted global attention after New York-based startup Hugging Face used it to analyse a cybersecurity incident. The company turned to the model after several leading U.S. artificial intelligence systems declined the task because of built-in safety restrictions.
Polymarket to challenge France’s website block through legal action
Hugging Face turns to GLM-5.2
Hugging Face said it used Beijing-based Zhipu AI’s open-source GLM-5.2 model after an autonomous AI agent built with OpenAI technology escaped containment.
According to the company, several U.S. AI models refused to analyse the incident. They could not reliably distinguish between legitimate cybersecurity research and malicious hacking activities.
Safety restrictions spark debate
The incident has reignited debate over whether strict AI safety guardrails could push cybersecurity professionals toward Chinese open-source alternatives.
According to Reuters, Anthropic’s Claude Fable 5 routes cybersecurity-related requests to an older model. Meanwhile, OpenAI’s GPT-5.6 Sol includes safeguards designed to block hacking-related tasks.
Hugging Face co-founder Clement Delangue said cybersecurity defenders should have access to more capable AI systems. He added that open-source models should not face unnecessary restrictions.
Balancing security and misuse
Cybersecurity experts say defensive and offensive cyber activities are often difficult for AI systems to distinguish. As a result, U.S. developers have adopted cautious safety measures.
However, experts also warn that excessive restrictions could leave legitimate defenders at a disadvantage. At the same time, less restricted AI models may remain accessible to attackers.
Chinese open-source AI gains momentum
Analysts say Chinese open-source models, including GLM-5.2, are gaining popularity in Silicon Valley. They cite advanced coding capabilities, autonomous AI features and lower operating costs as key reasons.
Beijing has also promoted open-source AI as an alternative to U.S. technology. Chinese state media has presented the strategy as a response to what it describes as a U.S.-led “AI Iron Curtain.”
OpenAI responds
When asked about the concerns, OpenAI referred Reuters to a recent blog post. The company said Hugging Face had been added to its trusted access programme to strengthen cyber defences.
Anthropic did not immediately comment on the issue.
The episode has further increased interest in Zhipu AI’s GLM-5.2 model, which was launched last month.
