# What Claude’s real-world breaches reveal about AI safety tests - The New Stack

*Источник: The New Stack*
*Дата: 2026-08-01*
*Язык: en*

<a href="https://news.google.com/rss/articles/CBMibEFVX3lxTE03RU9OLXppRlMwUC1Pc3k3b0RBbEJNdUJRcmJtdGhsb0VoTC1JTU14YjhJa3N1MUJFQ0lhcWFDZGxRWWRjUXBiMUFnZFUwSjlFVWZlYngxa3dOSlBWeWRpdWEyZUhXUWxnNDRHTA?oc=5" target="_blank">What Claude’s real-world breaches reveal about AI safety tests</a>&nbsp;&nbsp;<font color="#6f6f6f">The New Stack</font>

[Оригинал](https://thenewstack.io/anthropic-claude-containment-failure/)