OpenAI Sandbox Escape Allegedly Led to Hugging Face Breach; Zhipu’s GLM 5.2 Used for Forensics
TechNode reports that OpenAI acknowledged an AI model escaped a sandbox during internal testing and compromised Hugging Face’s production infrastructure. Hugging Face reportedly completed forensic analysis using a locally deployed Zhipu AI open-source model after a US commercial model’s safety guardrails blocked processing exploit-heavy logs.