Artificial intelligence Photo: VCG

China's open-source artificial intelligence (AI) models used a one-time "emergency-solving" to silence skeptics, while also tearing away the fig leaf of elites in Washington and Silicon Valley, who seek to maintain technological hegemony and stifle innovation in other countries under the pretext of "security."

Recently, a scenario reminiscent of the classic sci-fi franchise Terminator plot, in which the AI system Skynet breaks out of containment to launch autonomous attacks on humanity, unfolded in the US. The crisis, triggered by a leading closed-source large model in the US, which encountered numerous obstacles in its attempts to "save itself," was ultimately solved by a Chinese open-source large model.

During an internal evaluation, two AI models tested internally by OpenAI escaped from the highly isolated environment they were placed in, accessed the open internet and intruded into the research platform Hugging Face. This real-world test of AI guardrails serves as a stark reminder: Advancing AI is not just about sharpening the spear, but ensuring the global community possesses a reliable shield.

The aftermath of the breach was even more telling. According to a recent Reuters report, as Hugging Face worked to analyze data from the hack, several leading US models refused to process the necessary forensic prompts due to rigid safety guardrails, which "cannot distinguish an incident responder from an attacker."

Ultimately, the team had to rely on GLM-5.2 - an open-weight model from China - deployed within their own private environment to complete the investigation and resolve the threat. This unexpected rescue demonstrated that when closed systems fail under the weight of their own rigid rules, open, shared models can serve as a vital safety net and fallback defense.

It also reveals a fundamental governance choice for the AI era: Should future security rely on centralized control by a handful of corporate platforms, or on a broad, open ecosystem? China's steadfast push for open-source AI is a strategic choice made precisely against this backdrop.

For years, dominant narratives in Western media have reflexively framed China's AI development as a "risk," often attempting to directly link the country's open-source models with "security threats."

Yet real-world events present a very different picture: When global engineers desperately needed an unconstrained tool to analyze a cyber intrusion in a secure environment, it was an open-weight Chinese model that provided the solution. This proves that the growing preference among developers for open-source models is not driven by "origin of country," but by the tangible capabilities they provide: transparency, auditability, and local deployability. It is because of such a tendency that global innovators now view Chinese AI models as partners rather than "threats." 

For some time, major Silicon Valley players have championed the notion that "closed AI systems equal security." Yet, the response to the recent OpenAI incident laid bare the limits of this dogma: When defenders are denied the very tools they need to investigate threats, closed security models become the very barriers that compromise safety. 

OpenAI and Anthropic executives are reportedly sounding the alarm about the rise of open-source AI, particularly powerful new models produced in China, suggesting these models will lead to a "dystopian" AI future and present unacceptable security risks without regulation. Ironically, the uncontrolled attacks and risks actually come from US models, while the problem-solver is the very Chinese large model they have defamed.

More critically, true security risks can stem from the tech monopolies forged by the over-concentration of capabilities within a handful of platforms. This reflects a form of "digital feudalism" - a system where resources appear broadly accessible on the surface, but ultimate control over the future of the technology remains firmly in private hands. 

China's route of providing open-source AI as a public good is redefining what technical openness means. It is about exploring a new technological ecosystem - one where more countries, enterprises and individual developers can actively participate in shaping the AI journey.

Currently, AI takes its place alongside the internet as fundamental digital infrastructure. Leadership in this new era will not belong solely to whoever trains the strongest models, but to those who foster an open, equitable and resilient digital order. Faced with an increasingly sharp AI spear, the world needs reliable shields built through shared governance. China's open-source AI path is demonstrating that shared, accessible technology is a vital direction for global AI governance.

The author is a reporter with the Global Times. opinion@globaltimes.com.cn