Summary Hugging Face used a Chinese AI model after U.S. models declined cybersecurity tasks, highlighting concerns over restrictive AI safeguards.
NEW YORK (Reuters) – A Chinese artificial intelligence model has gained prominence after New York-based startup Hugging Face used it to analyse a cybersecurity incident when leading U.S. AI models declined the task due to built-in safety restrictions.
Hugging Face said it turned to 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. models refused to analyse the incident because they could not distinguish between defensive cybersecurity work and malicious hacking.
The incident has renewed debate over whether safety guardrails imposed by U.S. AI companies could push cybersecurity professionals towards Chinese open-source alternatives.
According to Reuters, Anthropic's Claude Fable 5 routes cybersecurity-related queries to an older model, while OpenAI's GPT-5.6 Sol includes protections designed to block hacking-related tasks.
Hugging Face co-founder Clement Delangue said all cybersecurity defenders should have access to more capable AI models without unnecessary restrictions, particularly open-source models.
Cybersecurity experts note that defensive and offensive cyber activities are often difficult for AI systems to distinguish, making U.S. developers cautious about relaxing safeguards. However, they warn that such restrictions could place legitimate defenders at a disadvantage while unrestricted models remain available to attackers.
Analysts say Chinese open-source models, including GLM-5.2, are rapidly gaining popularity in Silicon Valley because of their advanced coding and autonomous agent capabilities at comparatively lower cost. Beijing has also promoted open-source AI as an alternative to U.S. technology, with Chinese state media portraying the strategy as a response to what it describes as a U.S.-led "AI Iron Curtain."
When asked about the concerns, OpenAI referred Reuters to a recent blog post stating that Hugging Face had been added to its trusted access programme to help strengthen its cyber defences. Anthropic did not immediately comment.
The episode has further boosted the profile of Zhipu AI's GLM-5.2 model, which was launched last month.
