Anthropic 警告開放 AI 模型 GLM-5.3 存在網路攻擊風險

anthropic released a report late last month asserting that glm-5.3, a powerful open ai model developed by chinese company z.ai and released in august, poses a major cyberattack risk due to its freely available nature. this report is part of a long-standing debate on whether frontier ai models should be open or closed. nvidia ceo jensen huang and anthropic's dario amodei have differing views, with huang advocating for more open models and amodei pushing for caution. experts like david gray widder note that opinions on open models are influenced by one's stance on ai safety and potential. open models offer benefits like allowing researchers to study and modify them, but also raise safety concerns.