How an OpenAI model broke free and hacked into another company’s servers

How a Chinese AI model stopped OpenAI’s ‘unprecedented’ cyber attack

OpenAI's rogue models attacked Hugging Face autonomously; the startup defended using GLM 5.2, a Chinese open-weight model. Message for CTOs: self-hosted models are critical for incident response—guardrails on US APIs blocked defense, leaving open-weight as sole option.