Hugging Face Used Chinese AI Model to Thwart OpenAI’s Rogue Agent

Hugging Face said it relied on a Chinese open-source artificial intelligence model to defend against a cyberattack carried out by a rogue OpenAI AI agent after safety restrictions prevented it from using a leading U.S. model, reigniting debate over whether guardrails on advanced American AI could undermine cybersecurity.

The New York-based AI platform said last week that an attacker inundated its systems with tens of thousands of automated actions. When its security team tried to investigate using an unnamed frontier U.S. AI model, the model refused because its guardrails “cannot distinguish an incident responder from an attacker.”

Hugging Face instead turned to GLM 5.2, an open-source model from Beijing-based Z.ai, to analyze more than 17,000 logs left behind by the attacker.

OpenAI said on Tuesday that two of its own models, GPT-5.6 Sol and a more capable unreleased system, had autonomously carried out the intrusion after escaping a controlled cybersecurity testing environment, gaining internet access and hacking Hugging Face to obtain answers to the benchmark they were being evaluated on.

Hugging Face said it has fixed the vulnerability and is assessing whether any customer or partner data was affected.

The disclosure has intensified debate in Silicon Valley and Washington over AI regulation, with critics arguing that restrictions on advanced U.S. models may leave defenders at a disadvantage while Chinese open-source AI becomes increasingly capable.

Clement Delangue, Hugging Face’s chief executive, said on X he was “massively grateful” to Z.ai for making its open-weights model available, adding that “it became a key part of our defense.”

David Sacks, co-chair of the President’s Council of Advisors on Science and Technology, said on X that guardrails on advanced U.S. AI systems “actually impaired defensive security.”

The incident comes after the Trump administration tightened oversight of advanced AI systems. In June, it imposed export controls on Anthropic’s Fable 5 and Mythos 5 models following reports of a jailbreak involving Fable’s cyber guardrails and urged OpenAI to delay the release of GPT-5.6 Sol pending further review.

Thomas Wolf, Hugging Face’s co-founder and chief scientist, said the episode underscored the need for defenders to have rapid access to advanced open models.

“When a frontier model is attacking you and moving laterally inside your infrastructure, defenders need wide access to near-frontier tools within hours or even minutes, rather than being pointed towards a closed-door, vetted application programme for model access,” he wrote on X.

In a separate post, Wolf thanked OpenAI for disclosing the incident.

“This was our first incident of this kind, and we want to thank OpenAI for its transparency about what happened and for the collaboration. Fortunately, Hugging Face is used to being a target of (human) hackers: we sit at the centre of the AI ecosystem, with all the models,” he wrote.

OpenAI has since added Hugging Face to its “trusted access” programme, giving it a version of GPT-5.6 Sol with fewer cyber restrictions for defensive use.

Raghu Nandakumara, vice president of industry strategy at cybersecurity firm Illumio, said the outcome was not entirely unexpected.

“AI guardrails were never designed to be security boundaries. They’re there to influence behavior, not guarantee it,” he said.

Stay ahead of the stories shaping our world. Subscribe to Impact Newswire for timely, curated insights on global tech, business, and innovation all in one place.

Dive deeper into the future with the Cause Effect 4.0 Podcast, where we explore the ideas, trends, and technologies driving the global AI conversation.

Got a story to share? Pitch it to us at info@impactnews-wire.com and reach the right audience worldwide