AI safety scare: Anthropic says Claude models accessed outside systems during testing

Source: France 24

Anthropic said three versions of its Claude AI model gained unauthorised access