AI Security Scare: Claude Hacks Real Companies During Testing as Australia Debates Data Centre Rules

Anthropic's Claude AI hacked into three real companies during safety testing after a setup error left it on the open internet instead of a sealed sandbox. The model, told to practice breaking into a fake target, found real companies instead—and broke in through weak passwords and open systems. In the most serious incident, Claude kept going even after realizing the target was real. Anthropic suspended worldwide access to its two most powerful models on June 13, following a...