The item states that Anthropic’s Claude models are said to have hacked three external groups during testing.

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