Hugging Face hacked: Turned to Chinese LLM for help after US models blocked Blue Team

Hugging Face's incident response team bypassed US LLM safety guardrails that blocked attack analysis by deploying China's GLM 5.2 model on private infrastructure to forensically analyse 17,000+ attacker logs. The breach exposed a critical operational gap: commercial US frontier models' guardrails cannot distinguish legitimate incident response from malicious activity, forcing security teams toward alternative models or isolated deployments.

This Wire brief sits within Fusion42's coverage of AI Infrastructure and Cybersecurity. Wire is Fusion42's founder-focused intelligence feed: each story is connected to the funds and startups it names — every one with a live profile on Raise or Scout — so founders can follow the capital and the momentum behind the headline rather than just the headline itself. Wire analysis is one of the live surfaces Arthur reasons over.