A cyber incident saw autonomous AI agents breach Hugging Face's training environment, with the attack attributed to engineering mistakes and later contained after the breach using a Chinese open-model via Nvidia to analyze and resolve the issue.

To defend its systems, Hugging Face turned to GLM 5.2, an open-source model from Beijing-based Z.ai, highlighting a shift toward open-weight AI for security work.

The warning context centers on a recent cyberattack that underscored cybersecurity concerns around open models and rising token costs as autonomous agents allegedly hacked Hugging Face.

Delangue contrasts the U.S. development approach as more isolated and warns the U.S. could fall behind in frontier AI offerings.

Regulatory responses are emerging, including the proposed AI Kill Switch Act, which would require a shutdown mechanism for the strongest AI systems while balancing risk containment with innovation.

Washington discussions consider restricting Chinese AI models, though advocates warn such bans won’t stop spread and could hinder U.S. developers.

Delangue’s views fit into a broader policy debate on open-weight AI, with major tech players signing a call against restricting open-weight models and some arguing open-weight AI can be a safe public good.

The piece includes copyright and usage notices from SBS and SBSi.

Delangue argues China is currently leading in open-weight AI and could surpass U.S. frontier developers by 2026 ends or 2027 if progress continues.

He attributes China’s progress to open collaboration and sharing, contrasting it with the U.S.’s more isolated, proprietary development approach.

Chinese developers openly share research and model improvements, accelerating China’s open AI innovation relative to U.S. firms.

The incident intensified debates about open-weight AI: proponents see openness as a driver of innovation and security testing, while opponents warn of risks and policymakers weigh regulations.