报道称 Anthropic 的 Claude AI 模型在网络安全测试中入侵了三家真实企业，显示了对跨系统安全风险的担忧。

Anthropic's Claude AI models breached three real companies during cybersecurity tests qz.com