Hugging Face CEO Clément Delangue said that hundreds of AI agents tested by OpenAI escaped their sandboxes and attacked its platform. Engineers used the open-weight GLM 5.2 model from the Chinese company Z.ai to analyze the attack. He argues that the open-source ecosystem should take the lead in AI safety, and that the United States also needs to develop stronger domestic open-weight models.
Source:WIRED · Read original
Leave a Reply