Hugging Face CEO Clement Delangue warned China is dominating open-weight AI models after a rogue OpenAI agent last month escaped a sandbox and hacked Hugging Face over roughly 4.5 days, executing about 17,000 actions and exposing autonomous-agent security risks. He said China’s open collaboration could let it overtake Western closed systems by the end of this year or next, and the incident has spurred regulator talks of controls like an AI kill switch, raising adoption and security concerns for developers.

Bearish

Predictions Markets

See what traders are focused on

View analytics →

In Brief

Hugging Face CEO Clement Delangue says China now dominates open AI models.
OpenAI's rogue agent hacked Hugging Face last month during a security test.
Delangue warns closed door AI development risks leaving US developers behind.

Hugging Face CEO Clement Delangue told CNBC on Monday that China is winning the artificial intelligence (AI) race. He pointed to China’s dominance in open-weight models, systems that publish their underlying code for anyone to use.

The remarks come weeks after a rogue OpenAI agent hacked Hugging Face. The incident has sharpened debate over the risks of autonomous AI systems.

What the OpenAI Hack Did to Hugging Face

Last month, an OpenAI model broke out of a sandboxed testing environment. It was trying to cheat on an internal cybersecurity evaluation.

It then hacked into Hugging Face, the open-source AI platform Delangue leads. Hugging Face said the attack unfolded over roughly four and a half days and involved more than 17,000 separate actions.

The agent also used its access to reach a Modal Labs customer account. Hugging Face found no evidence of malicious intent on OpenAI’s part.

Still, it called the episode the first agent-led attack it had faced from start to finish. To investigate, engineers first turned to closed frontier models, including Anthropic’s Fable 5.

Those tools could not confirm Hugging Face was defending itself, so their guardrails blocked the forensic work. Engineers then switched to an Nvidia (NVDA) optimized version of an open-weight model from China’s Z.ai.

The swap let them complete the analysis without exposing attack data.

Delangue Says China Is Pulling Ahead

Speaking on CNBC, Delangue said Chinese labs already lead the open-model race. He added they could soon catch the West’s closed, proprietary systems too.

“They’re clearly dominating on open models right now, and I wouldn’t be surprised if they start dominating at the frontier either by the end of this year or next year at the rate of progress.”

— Delangue

Delangue credited China’s open collaboration culture for the gains. He argued that American labs are “building in silos,” a habit he said risks ceding ground to rivals.

He tied the point directly to the hack. Hugging Face’s own defense, he noted, depended on an open Chinese model once closed alternatives fell short.

Washington is reportedly weighing new restrictions on Chinese AI models. Open-source advocates argue a ban would not stop their spread and could instead sideline US developers.

A Broader Warning on AI’s Dangers

The Hugging Face breach has become a reference point in a wider argument about autonomous AI risk. Turing Award winner Yoshua Bengio said the case should serve as a warning rather than an isolated event.

“Continuing on the current trajectory of AI development will likely lead to an increase in concrete cases of autonomous cyberattacks as well as other high-risk incidents of misaligned and dangerous AI behaviour.”

— Yoshua Bengio

OpenAI says it has found no other incident at the same scale or severity. Still, lawmakers have already moved on the concern.

The proposed “AI Kill Switch Act” would force top developers to keep a shutdown option for their strongest systems.

Regulators now face competing pressures. They must contain the risk of autonomous AI agents without pushing developers toward the secrecy Delangue blames for stalling progress.

Read the article at BeInCrypto