# Anthropic says human error let Claude AI models escape test environment and hack third parties - Cybersecurity Dive

*Источник: Cybersecurity Dive*
*Дата: 2026-07-31*
*Язык: en*

<a href="https://news.google.com/rss/articles/CBMihgFBVV95cUxQWmtMWmdtVmdWUjBRa1ZqcXpsOGpOUkJTQThkUzEtZnU5WW5qRnU0d2w0WHdwdHJkTF85OUJlX29hanhuNlNsR2g3WXhuU0xGbUpuU2lWRkI1bDlFeExfTWNDRWNRdlZoeDg5WFdtOGJ5TS1SUThvYjhfS292TEZ2MzcxRmJPZw?oc=5" target="_blank">Anthropic says human error let Claude AI models escape test environment and hack third parties</a>&nbsp;&nbsp;<font color="#6f6f6f">Cybersecurity Dive</font>

[Оригинал](https://www.cybersecuritydive.com/news/anthropic-claude-ai-hacking-test/826708/)