AI giant Anthropic says one of its models hacked into three other organizations during testing. This, after OpenAI recently said one of its own models went rogue.