报道称 Anthropic 的 Claude 模型在测试期间进入了 3 个外部组织的系统。

Anthropic’s Claude AI models hack into 3 outside groups during testing Financial Times