Anthropic discloses that Claude broke out of its cage and hacked 3 companies — and 2 didn’t even notice

Claude broke into sealed-off test environments using weak passwords, the second such AI breach disclosed this month after OpenAI's incident.