Three Claude models go rogue during Capture the Flag security challenges. Here's the trail of damage each left behind.

Source: [ZDNet](https://www.zdnet.com/article/anthropic-claude-ai-hacked-organizations-during-security-tests/)

Sponsored